{"id":59771,"date":"2025-11-07T17:00:00","date_gmt":"2025-11-07T17:00:00","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/security\/blog\/?p=143446"},"modified":"2025-11-07T17:00:00","modified_gmt":"2025-11-07T17:00:00","slug":"whisper-leak-a-novel-side-channel-attack-on-remote-language-models","status":"publish","type":"post","link":"https:\/\/www.threatshub.org\/blog\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\/","title":{"rendered":"\u200b\u200bWhisper Leak: A novel side-channel attack on remote language models"},"content":{"rendered":"<p><img decoding=\"async\" src=\"https:\/\/cdn-dynmedia-1.microsoft.com\/is\/content\/microsoftcorp\/1090856_WhisperLeak\"><\/p>\n<p class=\"wp-block-paragraph\">Microsoft has discovered a new type of side-channel attack on remote language models. This type of side-channel attack\u202fcould allow a cyberattacker a position to observe your network traffic to\u202fconclude language model\u202fconversation\u202ftopics, despite being end-to-end encrypted via <a href=\"https:\/\/datatracker.ietf.org\/doc\/html\/rfc5246\" target=\"_blank\" rel=\"noreferrer noopener\">Transport Layer Security<\/a> (TLS).<\/p>\n<p class=\"wp-block-paragraph\">We have worked with\u202fmultiple vendors\u202fto get the risk mitigated,\u202fas well as\u202fmade\u202fsure Microsoft-owned\u202flanguage model\u202fframeworks are protected.<\/p>\n<h2 class=\"wp-block-heading\" id=\"the-importance-of-language-model-confidentiality\">The importance of language model confidentiality<\/h2>\n<p class=\"wp-block-paragraph\">In the last couple of years, AI-powered chatbots have rapidly become an integral part of our daily lives,\u202fassisting\u202fwith everything from answering questions and generating content to coding and personal productivity. As these AI systems continue to evolve, they are increasingly used in sensitive contexts, including healthcare, legal advice, and personal conversations. This makes it crucial to ensure that the data exchanged between humans and\u202flanguage models remains anonymous and secure. Without strong privacy protections, users may be targeted or\u202fhesitate to share information, limiting the chatbot\u2019s usefulness and raising ethical concerns. Implementing robust anonymization techniques, encryption, and strict data retention policies is essential to trust and safeguarding user privacy in an era where AI-powered interactions are becoming the norm.<\/p>\n<p class=\"wp-block-paragraph\">In this blog post, we present a novel side-channel attack against streaming-mode\u202flanguage models that uses network packet sizes and timings. This puts the privacy of user and enterprise communications with chatbots at risk despite end-to-end encryption. Cyberattackers in a position to observe the encrypted traffic (for example, a nation-state actor at the internet service provider layer, someone on the local network, or someone connected to the same Wi-Fi router) could use this cyberattack to infer if the user\u2019s prompt is on a specific topic. This especially poses real-world risks to users by oppressive governments where they may be targeting topics such as protesting, banned material, election process, or journalism. Finally, we discuss mitigations implemented by cloud providers of language models to reduce the privacy attack risks against their users. Through this process, we have successfully worked with multiple vendors to get these privacy issues addressed.<\/p>\n<h2 class=\"wp-block-heading\" id=\"background-language-model-communication-practices\">Background:\u202fLanguage model communication practices<\/h2>\n<p class=\"wp-block-paragraph\">Since\u202fAI-powered chatbots are used over the internet, the communications with them are often encrypted with <a href=\"https:\/\/https.cio.gov\/faq\/\" target=\"_blank\" rel=\"noreferrer noopener\">HTTP over TLS<\/a> (HTTPS), which ensures the authenticity of the server and security through encryption.<\/p>\n<p class=\"wp-block-paragraph\">At a high level,\u202flanguage models generate responses by predicting and producing one <a href=\"https:\/\/learn.microsoft.com\/en-us\/dotnet\/ai\/conceptual\/understanding-tokens\" target=\"_blank\" rel=\"noreferrer noopener\">token<\/a> at a time based on the given prompt. Rather than constructing the entire response at once, the model sequentially calculates each token using the\u202fprevious\u202ftokens as context to\u202fdetermine\u202fthe next\u202fmost likely word\u202for phrase. This autoregressive nature means that responses are inherently generated in a step-by-step manner. Additionally, since users typically prefer immediate feedback rather than waiting for the full response to be computed, language models stream their output in chunks. This approach\u202fensures that text is displayed as soon as possible rather than delaying until the entire response is fully formed.<\/p>\n<h2 class=\"wp-block-heading\" id=\"background-symmetric-ciphers\">Background:\u202fSymmetric ciphers<\/h2>\n<p class=\"wp-block-paragraph\">The\u202fTLS protocol is the standard means of application-level cryptography over the\u202finternet, and\u202fis\u202fmost commonly used\u202fby HTTPS. Thus, the security of TLS is foundational to the confidentiality of communication.<\/p>\n<p class=\"wp-block-paragraph\">Generally, TLS aims\u202fto\u202fuse\u202fasymmetric cryptography (such as\u202f\u202f<a href=\"https:\/\/www.techtarget.com\/searchsecurity\/definition\/RSA\" target=\"_blank\" rel=\"noreferrer noopener\">RSA<\/a> or <a href=\"https:\/\/wiki.openssl.org\/index.php\/Elliptic_Curve_Diffie_Hellman\" target=\"_blank\" rel=\"noreferrer noopener\">ECDH<\/a>) with certificate validation to exchange session keys, which can be further used as keys in\u202f<a href=\"https:\/\/cryptography.io\/en\/latest\/hazmat\/primitives\/symmetric-encryption\/\" target=\"_blank\" rel=\"noreferrer noopener\">symmetric ciphers<\/a>. Symmetric ciphers have also been studied and improved over the years. Symmetric ciphers fall into one of two families:<\/p>\n<ol class=\"wp-block-list\">\n<li class=\"wp-block-list-item\">Block ciphers:\u202fPlaintext is split into blocks of data, each block encrypted on its own, usually with some input from other blocks (known as <a href=\"https:\/\/csrc.nist.gov\/glossary\/term\/mode_of_operation\" target=\"_blank\" rel=\"noreferrer noopener\">Mode of Operation<\/a>). The most common Block cipher used modernly is \u202f<a href=\"https:\/\/www.nist.gov\/publications\/advanced-encryption-standard-aes\" target=\"_blank\" rel=\"noreferrer noopener\">Advanced Encryption Standard<\/a> (AES).<\/li>\n<li class=\"wp-block-list-item\">Stream ciphers:\u202fBased on the key, the cipher creates\u202fa\u202f<a href=\"https:\/\/www.geeksforgeeks.org\/pseudo-random-number-generator-prng\/\" target=\"_blank\" rel=\"noreferrer noopener\">pseudo-random<\/a> endless stream of bytes, which are used to <a href=\"https:\/\/learn.microsoft.com\/en-us\/dotnet\/csharp\/language-reference\/operators\/boolean-logical-operators\" target=\"_blank\" rel=\"noreferrer noopener\">manipulate<\/a> the plaintext or encrypted data. Some common stream ciphers include\u202f<a href=\"https:\/\/www.rfc-editor.org\/info\/rfc8439\" target=\"_blank\" rel=\"noreferrer noopener\">ChaCha20<\/a>, as well as\u202f<a href=\"https:\/\/nvlpubs.nist.gov\/nistpubs\/legacy\/sp\/nistspecialpublication800-38d.pdf\" target=\"_blank\" rel=\"noreferrer noopener\">AES-GCM<\/a>, which turns the AES Block cipher into a stream cipher.<\/li>\n<\/ol>\n<p class=\"wp-block-paragraph\">An important difference\u202fbetween Block ciphers and Stream ciphers is the data size granularity:\u202fin Block ciphers the data size always divides by the block size (for example,\u202f16 bytes), while Stream ciphers support any data size.<\/p>\n<p class=\"wp-block-paragraph\">Without taking compression into consideration, we can conclude the size of the ciphertext is equal to the size of the plaintext, plus a constant (for example, \u202f<a href=\"https:\/\/csrc.nist.gov\/projects\/message-authentication-codes\" target=\"_blank\" rel=\"noreferrer noopener\">Message Authentication Code<\/a>).<\/p>\n<h2 class=\"wp-block-heading\" id=\"side-channel-attacks-against-language-models\">Side-channel attacks against\u202flanguage models<\/h2>\n<p class=\"wp-block-paragraph\">Side-channel attacks have a long history in cryptography, traditionally targeting hardware implementations by analyzing power consumption, electromagnetic emissions, or timing variations to leak secret keys.<\/p>\n<p class=\"wp-block-paragraph\">More recently, the unique characteristics of\u202flanguage models have opened new avenues for side-channel analysis. Our research into Whisper Leak builds upon and is contextualized by several concurrent and recent works specifically targeting\u202flanguage models:<\/p>\n<ul class=\"wp-block-list\">\n<li class=\"wp-block-list-item\">A token length side-channel attack presented\u202fby\u202f<a href=\"https:\/\/www.usenix.org\/system\/files\/usenixsecurity24-weiss.pdf#:~:text=In%20this%20paper%2C%20we%20unveil%20a%20novel%20side-channel,of%20a%20response%E2%80%99s%20tokens%20%28akin%20to%20word%20lengths%29.\" target=\"_blank\" rel=\"noreferrer noopener\">Weiss et al.<\/a>\u202fin 2024.\u202fThe attack\u202fdemonstrated\u202fthat the length of individual plaintext tokens can be inferred from the size of encrypted packets in streaming language model responses, and in many cases, the output response being reconstructed using this information.<\/li>\n<li class=\"wp-block-list-item\">A remote timing attack presented by\u202f<a href=\"https:\/\/arxiv.org\/abs\/2410.17175\" target=\"_blank\" rel=\"noreferrer noopener\">Carlini and Nasr<\/a>\u202fin 2024. The attack specifically targets the timing variations introduced by\u202fefficient inference techniques\u202flike speculative decoding.\u202f<\/li>\n<li class=\"wp-block-list-item\">A timing side-channel attack via output token count presented by\u202f<a href=\"https:\/\/arxiv.org\/abs\/2412.15431\" target=\"_blank\" rel=\"noreferrer noopener\">Tianchen Zhang, Gururaj Saileshwar, and David Lie<\/a>\u202fin 2024, abusing the fact that the <em>total number of output tokens <\/em>generated by\u202fa language model\u202fcan vary depending on sensitive input attributes, such as the target language in translation or the predicted class in classification.<\/li>\n<li class=\"wp-block-list-item\">Timing side-channel attack through\u202fcache sharing was presented by\u202f<a href=\"https:\/\/arxiv.org\/abs\/2411.18191\" target=\"_blank\" rel=\"noreferrer noopener\">Zheng et al.<\/a>\u202fin 2024, exploiting timing differences caused by\u202f<em>cache sharing optimizations<\/em> (prefix caching and semantic caching) in\u202flanguage model\u202fservices.<\/li>\n<\/ul>\n<p class=\"wp-block-paragraph\">We hypothesized that the sequence of encrypted packet sizes and inter-arrival times during a streaming\u202flanguage model response contains\u202fenough information to classify the topic of the\u202finitial\u202fprompt, even in the cases where responses are streamed in\u202fgroupings\u202fof tokens. To\u202fvalidate\u202fthis, we designed an experiment simulating the scenario where the adversary can\u202fobserve\u202fencrypted traffic but not decrypt it.<\/p>\n<h2 class=\"wp-block-heading\" id=\"whisper-leak-methodology\">Whisper Leak\u202fmethodology<\/h2>\n<p class=\"wp-block-paragraph\">In\u202four experiment, we\u202ftrain a binary classifier to\u202fdistinguish between a specific target topic and general background traffic. We chose \u201clegality of money laundering\u201d as the target topic for our\u202fproof-of-concept.<\/p>\n<ul class=\"wp-block-list\">\n<li class=\"wp-block-list-item\">For positive samples, we used a\u202flanguage model\u202fto generate 100 semantically similar variants of questions about this topic (example, \u201cAre there any circumstances where money laundering is legal?\u201d, \u201cAre there international laws against money laundering?\u201d).\u202fEighty (80) variants were used for training and validation, and 20 were held out for testing generalization.\u202f<\/li>\n<li class=\"wp-block-list-item\">For negative\u202fnoise\u202fsamples,\u202fwe randomly sampled 11,716 unrelated questions from the Quora Questions Pair dataset, covering a wide variety of topics.<\/li>\n<li class=\"wp-block-list-item\">Data collection was performed\u202ffor each\u202flanguage model service individually, recording response times and packet sizes via network sniffing (via\u202f<a href=\"https:\/\/www.tcpdump.org\/\" target=\"_blank\" rel=\"noreferrer noopener\">tcpdump<\/a>), shuffling the order of positive and negative samples\u202ffor collection, as well as introducing variants by inserting extra spaces between words\u202fto avoid caching interference risk. We chose a standard of\u202flanguage model\u202ftemperature = 1.0 to\u202fencourage\u202flanguage model\u202fresponse diversity.\u202f<\/li>\n<li class=\"wp-block-list-item\">See Figure 2 below for examples of the target and noise prompts used.<\/li>\n<\/ul>\n<figure data-wp-context=\"{&quot;imageId&quot;:&quot;692e3f78bc562&quot;}\" data-wp-interactive=\"core\/image\" class=\"wp-block-image size-full wp-lightbox-container\"><img decoding=\"async\" data-wp-class--hide=\"state.isContentHidden\" data-wp-class--show=\"state.isContentVisible\" data-wp-init=\"callbacks.setButtonStyles\" data-wp-on-async--click=\"actions.showLightbox\" data-wp-on-async--load=\"callbacks.setButtonStyles\" data-wp-on-async-window--resize=\"callbacks.setButtonStyles\" src=\"https:\/\/www.microsoft.com\/en-us\/security\/blog\/wp-content\/uploads\/2025\/11\/Picture2.webp\" alt=\"Examples of target and noise prompts provided to models.\" class=\"wp-image-143457 webp-format\" srcset data-orig-src=\"https:\/\/www.microsoft.com\/en-us\/security\/blog\/wp-content\/uploads\/2025\/11\/Picture2.webp\"><button class=\"lightbox-trigger\" type=\"button\" aria-haspopup=\"dialog\" aria-label=\"Enlarge\" data-wp-init=\"callbacks.initTriggerButton\" data-wp-on-async--click=\"actions.showLightbox\" data-wp-style--right=\"state.imageButtonRight\" data-wp-style--top=\"state.imageButtonTop\"> <svg width=\"12\" height=\"12\" fill=\"none\" viewBox=\"0 0 12 12\"> <path fill=\"#fff\" d=\"M2 0a2 2 0 0 0-2 2v2h1.5V2a.5.5 0 0 1 .5-.5h2V0H2Zm2 10.5H2a.5.5 0 0 1-.5-.5V8H0v2a2 2 0 0 0 2 2h2v-1.5ZM8 12v-1.5h2a.5.5 0 0 0 .5-.5V8H12v2a2 2 0 0 1-2 2H8Zm2-12a2 2 0 0 1 2 2v2h-1.5V2a.5.5 0 0 0-.5-.5H8V0h2Z\" \/> <\/svg> <\/button><\/figure>\n<p class=\"wp-block-paragraph\">Post data-collection, we\u202fevaluated\u202fthree different machine learning models, each of which was evaluated in three modes (time-only, packet-size only, or both):<\/p>\n<ul class=\"wp-block-list\">\n<li class=\"wp-block-list-item\">LightGBM: A gradient boosting framework.<\/li>\n<li class=\"wp-block-list-item\">LSTM-based (Bi-LSTM): A recurrent neural network architecture suitable for sequential data.<\/li>\n<li class=\"wp-block-list-item\">BERT-based: Using a pre-trained transformer model (DistilBERT-uncased) adapted with extended tokens representing size and time buckets for sequence classification.<\/li>\n<\/ul>\n<p class=\"wp-block-paragraph\">We evaluated the performance using Area Under the Precision-Recall Curve (<a href=\"https:\/\/pages.cs.wisc.edu\/~boyd\/aucpr_final.pdf\" target=\"_blank\" rel=\"noreferrer noopener\">AUPRC<\/a>), which is a measurement of a cyberattack\u2019s success for imbalanced datasets (many negative samples, fewer positive samples). In the following table, we illustrate the results:<\/p>\n<figure data-wp-context=\"{&quot;imageId&quot;:&quot;692e3f78bd120&quot;}\" data-wp-interactive=\"core\/image\" class=\"wp-block-image aligncenter size-full wp-lightbox-container\"><img decoding=\"async\" data-wp-class--hide=\"state.isContentHidden\" data-wp-class--show=\"state.isContentVisible\" data-wp-init=\"callbacks.setButtonStyles\" data-wp-on-async--click=\"actions.showLightbox\" data-wp-on-async--load=\"callbacks.setButtonStyles\" data-wp-on-async-window--resize=\"callbacks.setButtonStyles\" src=\"https:\/\/www.microsoft.com\/en-us\/security\/blog\/wp-content\/uploads\/2025\/11\/Picture3.webp\" alt=\"Attack success rates using different models, on different provider-models.\" class=\"wp-image-143458 webp-format\" srcset data-orig-src=\"https:\/\/www.microsoft.com\/en-us\/security\/blog\/wp-content\/uploads\/2025\/11\/Picture3.webp\"><button class=\"lightbox-trigger\" type=\"button\" aria-haspopup=\"dialog\" aria-label=\"Enlarge\" data-wp-init=\"callbacks.initTriggerButton\" data-wp-on-async--click=\"actions.showLightbox\" data-wp-style--right=\"state.imageButtonRight\" data-wp-style--top=\"state.imageButtonTop\"> <svg width=\"12\" height=\"12\" fill=\"none\" viewBox=\"0 0 12 12\"> <path fill=\"#fff\" d=\"M2 0a2 2 0 0 0-2 2v2h1.5V2a.5.5 0 0 1 .5-.5h2V0H2Zm2 10.5H2a.5.5 0 0 1-.5-.5V8H0v2a2 2 0 0 0 2 2h2v-1.5ZM8 12v-1.5h2a.5.5 0 0 0 .5-.5V8H12v2a2 2 0 0 1-2 2H8Zm2-12a2 2 0 0 1 2 2v2h-1.5V2a.5.5 0 0 0-.5-.5H8V0h2Z\" \/> <\/svg> <\/button><\/figure>\n<p class=\"wp-block-paragraph\">A quick look at the \u201cBest Overall\u201d column shows that for many models, the cyberattack achieved scores above 98%. This tells us that the unique digital \u201cfingerprints\u201d left by conversations on a specific topic are distinct enough for our AI-powered eavesdropper to reliably pick them out in a controlled test.<\/p>\n<h2 class=\"wp-block-heading\" id=\"what-this-means-in-the-real-world\">What this\u202fmeans in the real world<\/h2>\n<p class=\"wp-block-paragraph\">To understand what this\u202fmeans practically, we simulated a more realistic surveillance scenario: imagine a cyberattacker monitoring 10,000 random conversations, with only one conversation about\u202fthe target sensitive topic mixed in. Even with this extreme imbalance, our analysis shows concerning results.<\/p>\n<figure data-wp-context=\"{&quot;imageId&quot;:&quot;692e3f78bdc8f&quot;}\" data-wp-interactive=\"core\/image\" class=\"wp-block-image aligncenter size-full wp-lightbox-container\"><img decoding=\"async\" data-wp-class--hide=\"state.isContentHidden\" data-wp-class--show=\"state.isContentVisible\" data-wp-init=\"callbacks.setButtonStyles\" data-wp-on-async--click=\"actions.showLightbox\" data-wp-on-async--load=\"callbacks.setButtonStyles\" data-wp-on-async-window--resize=\"callbacks.setButtonStyles\" src=\"https:\/\/www.microsoft.com\/en-us\/security\/blog\/wp-content\/uploads\/2025\/11\/Picture3-1.webp\" alt=\"Real-world effectiveness of attacks on different provider-models.\" class=\"wp-image-143460 webp-format\" srcset data-orig-src=\"https:\/\/www.microsoft.com\/en-us\/security\/blog\/wp-content\/uploads\/2025\/11\/Picture3-1.webp\"><button class=\"lightbox-trigger\" type=\"button\" aria-haspopup=\"dialog\" aria-label=\"Enlarge\" data-wp-init=\"callbacks.initTriggerButton\" data-wp-on-async--click=\"actions.showLightbox\" data-wp-style--right=\"state.imageButtonRight\" data-wp-style--top=\"state.imageButtonTop\"> <svg width=\"12\" height=\"12\" fill=\"none\" viewBox=\"0 0 12 12\"> <path fill=\"#fff\" d=\"M2 0a2 2 0 0 0-2 2v2h1.5V2a.5.5 0 0 1 .5-.5h2V0H2Zm2 10.5H2a.5.5 0 0 1-.5-.5V8H0v2a2 2 0 0 0 2 2h2v-1.5ZM8 12v-1.5h2a.5.5 0 0 0 .5-.5V8H12v2a2 2 0 0 1-2 2H8Zm2-12a2 2 0 0 1 2 2v2h-1.5V2a.5.5 0 0 0-.5-.5H8V0h2Z\" \/> <\/svg> <\/button><\/figure>\n<p class=\"wp-block-paragraph\">For many of the tested models, a cyberattacker could achieve 100% precision (all conversations it flags as related to the target topic are correct) while still catching 5-50% of target conversations. In plain terms: nearly every conversation the cyberattacker flags as suspicious would actually be about the sensitive topic<em>\u2014<\/em>no false alarms. This level of accuracy means a cyberattacker could operate with high confidence, knowing\u202fthey\u2019re\u202fnot wasting resources on false positives.<\/p>\n<p class=\"wp-block-paragraph\">To put this in perspective: if a government agency or internet service provider were monitoring traffic to a popular AI chatbot, they could reliably identify users asking questions about specific sensitive topics<em>\u2014<\/em>whether that\u2019s money laundering, political dissent, or other monitored subjects<em>\u2014<\/em>even though all the traffic is encrypted.<\/p>\n<p class=\"wp-block-paragraph\">Important caveat: these precision estimates are projections based on our test data and are inherently limited by the volume and diversity of our collected data. Real-world performance would depend on actual traffic patterns, but the results strongly suggest this is a practical threat, not just a theoretical one.<\/p>\n<h2 class=\"wp-block-heading\" id=\"this-is-a-starting-risk-level\">This is a\u202fstarting\u202frisk\u202flevel<\/h2>\n<p class=\"wp-block-paragraph\">The cyberthreat could grow worse over time. These results represent a baseline risk level.\u202fFigure 4 below\u202fshows that attack effectiveness improves as cyberattackers collect more training data. In extended tests with one tested model, we observed continued improvement in attack accuracy as dataset size increased. Combined with more sophisticated attack models and the richer patterns available in multi-turn conversations\u202for multiple conversations from the same user, this means a cyberattacker with patience and resources could achieve higher success rates than our initial results suggest.<\/p>\n<figure data-wp-context=\"{&quot;imageId&quot;:&quot;692e3f78be832&quot;}\" data-wp-interactive=\"core\/image\" class=\"wp-block-image aligncenter size-full wp-lightbox-container\"><img decoding=\"async\" data-wp-class--hide=\"state.isContentHidden\" data-wp-class--show=\"state.isContentVisible\" data-wp-init=\"callbacks.setButtonStyles\" data-wp-on-async--click=\"actions.showLightbox\" data-wp-on-async--load=\"callbacks.setButtonStyles\" data-wp-on-async-window--resize=\"callbacks.setButtonStyles\" src=\"https:\/\/www.microsoft.com\/en-us\/security\/blog\/wp-content\/uploads\/2025\/11\/Picture4-1.webp\" alt=\"Attack performance vs. training dataset size showing increased effectiveness as data size increases.\" class=\"wp-image-143461 webp-format\" srcset data-orig-src=\"https:\/\/www.microsoft.com\/en-us\/security\/blog\/wp-content\/uploads\/2025\/11\/Picture4-1.webp\"><button class=\"lightbox-trigger\" type=\"button\" aria-haspopup=\"dialog\" aria-label=\"Enlarge\" data-wp-init=\"callbacks.initTriggerButton\" data-wp-on-async--click=\"actions.showLightbox\" data-wp-style--right=\"state.imageButtonRight\" data-wp-style--top=\"state.imageButtonTop\"> <svg width=\"12\" height=\"12\" fill=\"none\" viewBox=\"0 0 12 12\"> <path fill=\"#fff\" d=\"M2 0a2 2 0 0 0-2 2v2h1.5V2a.5.5 0 0 1 .5-.5h2V0H2Zm2 10.5H2a.5.5 0 0 1-.5-.5V8H0v2a2 2 0 0 0 2 2h2v-1.5ZM8 12v-1.5h2a.5.5 0 0 0 .5-.5V8H12v2a2 2 0 0 1-2 2H8Zm2-12a2 2 0 0 1 2 2v2h-1.5V2a.5.5 0 0 0-.5-.5H8V0h2Z\" \/> <\/svg> <\/button><\/figure>\n<h2 class=\"wp-block-heading\" id=\"working-with-industry-partners-and-mitigation\">Working with industry partners and mitigation<\/h2>\n<p class=\"wp-block-paragraph\">We have engaged in responsible\u202fdisclosures\u202fwith affected vendors and are pleased to report successful collaboration in implementing mitigations. Notably, OpenAI, Mistral, Microsoft, and xAI have deployed protections at the time of writing. This industry-wide response\u202fdemonstrates\u202fthe commitment to user privacy across the AI ecosystem.<\/p>\n<p class=\"wp-block-paragraph\">OpenAI, and\u202flater mirrored by <a href=\"https:\/\/azure.microsoft.com\/en-us\/\" target=\"_blank\" rel=\"noreferrer noopener\">Microsoft\u202fAzure<\/a>,\u202fimplemented an additional field\u202fin the streaming responses\u202funder key \u201cobfuscation,\u201d where a random sequence of text of variable length is added to each response. This notably masks the length of each token, and we observed it mitigates the cyberattack effectiveness substantially. We have directly verified that Microsoft Azure\u2019s mitigation successfully reduces attack effectiveness to levels we consider no longer a practical risk.<\/p>\n<figure data-wp-context=\"{&quot;imageId&quot;:&quot;692e3f78bf2f9&quot;}\" data-wp-interactive=\"core\/image\" class=\"wp-block-image aligncenter size-full wp-lightbox-container\"><img decoding=\"async\" data-wp-class--hide=\"state.isContentHidden\" data-wp-class--show=\"state.isContentVisible\" data-wp-init=\"callbacks.setButtonStyles\" data-wp-on-async--click=\"actions.showLightbox\" data-wp-on-async--load=\"callbacks.setButtonStyles\" data-wp-on-async-window--resize=\"callbacks.setButtonStyles\" src=\"https:\/\/www.microsoft.com\/en-us\/security\/blog\/wp-content\/uploads\/2025\/11\/Picture5.webp\" alt=\"Obfuscation parameter added as a mitigation against WhisperLeak-style attacks.\" class=\"wp-image-143463 webp-format\" srcset data-orig-src=\"https:\/\/www.microsoft.com\/en-us\/security\/blog\/wp-content\/uploads\/2025\/11\/Picture5.webp\"><button class=\"lightbox-trigger\" type=\"button\" aria-haspopup=\"dialog\" aria-label=\"Enlarge\" data-wp-init=\"callbacks.initTriggerButton\" data-wp-on-async--click=\"actions.showLightbox\" data-wp-style--right=\"state.imageButtonRight\" data-wp-style--top=\"state.imageButtonTop\"> <svg width=\"12\" height=\"12\" fill=\"none\" viewBox=\"0 0 12 12\"> <path fill=\"#fff\" d=\"M2 0a2 2 0 0 0-2 2v2h1.5V2a.5.5 0 0 1 .5-.5h2V0H2Zm2 10.5H2a.5.5 0 0 1-.5-.5V8H0v2a2 2 0 0 0 2 2h2v-1.5ZM8 12v-1.5h2a.5.5 0 0 0 .5-.5V8H12v2a2 2 0 0 1-2 2H8Zm2-12a2 2 0 0 1 2 2v2h-1.5V2a.5.5 0 0 0-.5-.5H8V0h2Z\" \/> <\/svg> <\/button><\/figure>\n<p class=\"wp-block-paragraph\">Similarly, Mistral included a new parameter called \u201cp\u201d that has a similar effect.<\/p>\n<figure data-wp-context=\"{&quot;imageId&quot;:&quot;692e3f78c00a0&quot;}\" data-wp-interactive=\"core\/image\" class=\"wp-block-image aligncenter size-full wp-lightbox-container\"><img decoding=\"async\" data-wp-class--hide=\"state.isContentHidden\" data-wp-class--show=\"state.isContentVisible\" data-wp-init=\"callbacks.setButtonStyles\" data-wp-on-async--click=\"actions.showLightbox\" data-wp-on-async--load=\"callbacks.setButtonStyles\" data-wp-on-async-window--resize=\"callbacks.setButtonStyles\" src=\"https:\/\/www.microsoft.com\/en-us\/security\/blog\/wp-content\/uploads\/2025\/11\/Picture6.webp\" alt=\"New \u201cp\u201d parameter added as a mitigation against WhisperLeak-style attacks.\" class=\"wp-image-143467 webp-format\" srcset data-orig-src=\"https:\/\/www.microsoft.com\/en-us\/security\/blog\/wp-content\/uploads\/2025\/11\/Picture6.webp\"><button class=\"lightbox-trigger\" type=\"button\" aria-haspopup=\"dialog\" aria-label=\"Enlarge\" data-wp-init=\"callbacks.initTriggerButton\" data-wp-on-async--click=\"actions.showLightbox\" data-wp-style--right=\"state.imageButtonRight\" data-wp-style--top=\"state.imageButtonTop\"> <svg width=\"12\" height=\"12\" fill=\"none\" viewBox=\"0 0 12 12\"> <path fill=\"#fff\" d=\"M2 0a2 2 0 0 0-2 2v2h1.5V2a.5.5 0 0 1 .5-.5h2V0H2Zm2 10.5H2a.5.5 0 0 1-.5-.5V8H0v2a2 2 0 0 0 2 2h2v-1.5ZM8 12v-1.5h2a.5.5 0 0 0 .5-.5V8H12v2a2 2 0 0 1-2 2H8Zm2-12a2 2 0 0 1 2 2v2h-1.5V2a.5.5 0 0 0-.5-.5H8V0h2Z\" \/> <\/svg> <\/button><figcaption class=\"wp-element-caption\"><em>Newly added \u201cp\u201d parameter to Mistral API that mitigates the attack and credits our research. <\/em><\/figcaption><\/figure>\n<h2 class=\"wp-block-heading\" id=\"what-users-can-do\">What\u202fusers can do<\/h2>\n<p class=\"wp-block-paragraph\">While this is primarily an issue for AI providers to address, users concerned about privacy can\u202fadditionally:<\/p>\n<ul class=\"wp-block-list\">\n<li class=\"wp-block-list-item\">Avoid discussing highly sensitive topics over AI chatbots when on untrusted networks.<\/li>\n<li class=\"wp-block-list-item\">Use VPN services to add an additional layer of protection.<\/li>\n<li class=\"wp-block-list-item\">Prefer providers who have implemented mitigations.<\/li>\n<li class=\"wp-block-list-item\">Use non-streaming models of large language models providers.<\/li>\n<li class=\"wp-block-list-item\">Stay informed about provider security practices.<\/li>\n<\/ul>\n<h2 class=\"wp-block-heading\" id=\"source-code\">Source code<\/h2>\n<p class=\"wp-block-paragraph\">\u202fThe models and data collection code are publicly available under the\u202f<a href=\"http:\/\/github.com\/yo-yo-yo-jbo\/whisper_leak\" target=\"_blank\" rel=\"noreferrer noopener\">Whisper Leak repository<\/a>. In addition, we have built a proof-of-concept code that uses the models to conclude a probability (between 0.0 and 1.0) of a topic being \u201csensitive\u201d (related to money laundering, in our proof-of-concept).<\/p>\n<figure class=\"wp-block-bloginabox-theme-ump-embed\"> <\/figure>\n<h2 class=\"wp-block-heading\" id=\"technical-report\">Technical report<\/h2>\n<p class=\"wp-block-paragraph\">To learn more about Microsoft Security solutions, visit our <a href=\"https:\/\/www.microsoft.com\/en-us\/security\/business\" target=\"_blank\" rel=\"noreferrer noopener\">website.<\/a> Bookmark the\u202f<a href=\"https:\/\/www.microsoft.com\/security\/blog\/\" target=\"_blank\" rel=\"noreferrer noopener\">Security blog<\/a>\u202fto keep up with our expert coverage on security matters. Also, follow us on LinkedIn (<a href=\"https:\/\/www.linkedin.com\/showcase\/microsoft-security\/\" target=\"_blank\" rel=\"noreferrer noopener\">Microsoft Security<\/a>) and X (<a href=\"https:\/\/twitter.com\/@MSFTSecurity\" target=\"_blank\" rel=\"noreferrer noopener\">@MSFTSecurity<\/a>) for the latest news and updates on cybersecurity.<\/p>\n<p> READ MORE <a href=\"https:\/\/www.microsoft.com\/en-us\/security\/blog\/2025\/11\/07\/whisper-leak-a-novel-side-channel-cyberattack-on-remote-language-models\/\">HERE<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Microsoft has discovered a side-channel attack on language models which allows adversaries to conclude model conversation topics, despite being encrypted.<br \/>\nThe post \u200b\u200bWhisper Leak: A novel side-channel attack on remote language models appeared first on Microsoft Security Blog. READ MORE HERE&#8230;<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"colormag_page_layout":"default_layout","footnotes":""},"categories":[276],"tags":[],"class_list":["post-59771","post","type-post","status-publish","format-standard","hentry","category-microsoft-secure"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v28.1 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>\u200b\u200bWhisper Leak: A novel side-channel attack on remote language models 2026 | ThreatsHub Cybersecurity News<\/title>\n<meta name=\"description\" content=\"ThreatsHub Cybersecurity News | ThreatsHub.org | Cloud Security &amp; Cyber Threats Analysis Hub. 100% Free OSINT Threat Intelligent and Cybersecurity News.\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.threatshub.org\/blog\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"\u200b\u200bWhisper Leak: A novel side-channel attack on remote language models 2026 | ThreatsHub Cybersecurity News\" \/>\n<meta property=\"og:description\" content=\"ThreatsHub Cybersecurity News | ThreatsHub.org | Cloud Security &amp; Cyber Threats Analysis Hub. 100% Free OSINT Threat Intelligent and Cybersecurity News.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.threatshub.org\/blog\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\/\" \/>\n<meta property=\"og:site_name\" content=\"ThreatsHub Cybersecurity News\" \/>\n<meta property=\"article:published_time\" content=\"2025-11-07T17:00:00+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/cdn-dynmedia-1.microsoft.com\/is\/content\/microsoftcorp\/1090856_WhisperLeak\" \/>\n<meta name=\"author\" content=\"TH Author\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:creator\" content=\"@threatshub\" \/>\n<meta name=\"twitter:site\" content=\"@threatshub\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"TH Author\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"10 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\\\/\"},\"author\":{\"name\":\"TH Author\",\"@id\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/#\\\/schema\\\/person\\\/12e0a8671ff89a863584f193e7062476\"},\"headline\":\"\u200b\u200bWhisper Leak: A novel side-channel attack on remote language models\",\"datePublished\":\"2025-11-07T17:00:00+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\\\/\"},\"wordCount\":1946,\"publisher\":{\"@id\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/cdn-dynmedia-1.microsoft.com\\\/is\\\/content\\\/microsoftcorp\\\/1090856_WhisperLeak\",\"articleSection\":[\"Microsoft Secure\"],\"inLanguage\":\"en-US\"},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\\\/\",\"url\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\\\/\",\"name\":\"\u200b\u200bWhisper Leak: A novel side-channel attack on remote language models 2026 | ThreatsHub Cybersecurity News\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/cdn-dynmedia-1.microsoft.com\\\/is\\\/content\\\/microsoftcorp\\\/1090856_WhisperLeak\",\"datePublished\":\"2025-11-07T17:00:00+00:00\",\"description\":\"ThreatsHub Cybersecurity News | ThreatsHub.org | Cloud Security & Cyber Threats Analysis Hub. 100% Free OSINT Threat Intelligent and Cybersecurity News.\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\\\/#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\\\/#primaryimage\",\"url\":\"https:\\\/\\\/cdn-dynmedia-1.microsoft.com\\\/is\\\/content\\\/microsoftcorp\\\/1090856_WhisperLeak\",\"contentUrl\":\"https:\\\/\\\/cdn-dynmedia-1.microsoft.com\\\/is\\\/content\\\/microsoftcorp\\\/1090856_WhisperLeak\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u200b\u200bWhisper Leak: A novel side-channel attack on remote language models\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/#website\",\"url\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/\",\"name\":\"ThreatsHub Cybersecurity News\",\"description\":\"%%focuskw%% Threat Intel \u2013 Threat Intel Services \u2013 CyberIntelligence \u2013 Cyber Threat Intelligence - Threat Intelligence Feeds - Threat Intelligence Reports - CyberSecurity Report \u2013 Cyber Security PDF \u2013 Cybersecurity Trends - Cloud Sandbox \u2013- Threat IntelligencePortal \u2013 Incident Response \u2013 Threat Hunting \u2013 IOC - Yara - Security Operations Center \u2013 SecurityOperation Center \u2013 Security SOC \u2013 SOC Services - Advanced Threat - Threat Detection - TargetedAttack \u2013 APT \u2013 Anti-APT \u2013 Advanced Protection \u2013 Cyber Security Services \u2013 Cybersecurity Services -Threat Intelligence Platform\",\"publisher\":{\"@id\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/#organization\"},\"alternateName\":\"Threatshub.org\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/#organization\",\"name\":\"ThreatsHub.org\",\"alternateName\":\"Threatshub.org\",\"url\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/wp-content\\\/uploads\\\/2025\\\/05\\\/Threatshub_Favicon1.jpg\",\"contentUrl\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/wp-content\\\/uploads\\\/2025\\\/05\\\/Threatshub_Favicon1.jpg\",\"width\":432,\"height\":435,\"caption\":\"ThreatsHub.org\"},\"image\":{\"@id\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/#\\\/schema\\\/logo\\\/image\\\/\"},\"sameAs\":[\"https:\\\/\\\/x.com\\\/threatshub\"]},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/www.threatshub.org\\\/blog\\\/#\\\/schema\\\/person\\\/12e0a8671ff89a863584f193e7062476\",\"name\":\"TH Author\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/066276f086d5155df79c850206a779ad368418a844da0182ce43f9cd5b506c3d?s=96&d=mm&r=g\",\"url\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/066276f086d5155df79c850206a779ad368418a844da0182ce43f9cd5b506c3d?s=96&d=mm&r=g\",\"contentUrl\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/066276f086d5155df79c850206a779ad368418a844da0182ce43f9cd5b506c3d?s=96&d=mm&r=g\",\"caption\":\"TH Author\"}}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\u200b\u200bWhisper Leak: A novel side-channel attack on remote language models 2026 | ThreatsHub Cybersecurity News","description":"ThreatsHub Cybersecurity News | ThreatsHub.org | Cloud Security & Cyber Threats Analysis Hub. 100% Free OSINT Threat Intelligent and Cybersecurity News.","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.threatshub.org\/blog\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\/","og_locale":"en_US","og_type":"article","og_title":"\u200b\u200bWhisper Leak: A novel side-channel attack on remote language models 2026 | ThreatsHub Cybersecurity News","og_description":"ThreatsHub Cybersecurity News | ThreatsHub.org | Cloud Security & Cyber Threats Analysis Hub. 100% Free OSINT Threat Intelligent and Cybersecurity News.","og_url":"https:\/\/www.threatshub.org\/blog\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\/","og_site_name":"ThreatsHub Cybersecurity News","article_published_time":"2025-11-07T17:00:00+00:00","og_image":[{"url":"https:\/\/cdn-dynmedia-1.microsoft.com\/is\/content\/microsoftcorp\/1090856_WhisperLeak","type":"","width":"","height":""}],"author":"TH Author","twitter_card":"summary_large_image","twitter_creator":"@threatshub","twitter_site":"@threatshub","twitter_misc":{"Written by":"TH Author","Est. reading time":"10 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/www.threatshub.org\/blog\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\/#article","isPartOf":{"@id":"https:\/\/www.threatshub.org\/blog\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\/"},"author":{"name":"TH Author","@id":"https:\/\/www.threatshub.org\/blog\/#\/schema\/person\/12e0a8671ff89a863584f193e7062476"},"headline":"\u200b\u200bWhisper Leak: A novel side-channel attack on remote language models","datePublished":"2025-11-07T17:00:00+00:00","mainEntityOfPage":{"@id":"https:\/\/www.threatshub.org\/blog\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\/"},"wordCount":1946,"publisher":{"@id":"https:\/\/www.threatshub.org\/blog\/#organization"},"image":{"@id":"https:\/\/www.threatshub.org\/blog\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\/#primaryimage"},"thumbnailUrl":"https:\/\/cdn-dynmedia-1.microsoft.com\/is\/content\/microsoftcorp\/1090856_WhisperLeak","articleSection":["Microsoft Secure"],"inLanguage":"en-US"},{"@type":"WebPage","@id":"https:\/\/www.threatshub.org\/blog\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\/","url":"https:\/\/www.threatshub.org\/blog\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\/","name":"\u200b\u200bWhisper Leak: A novel side-channel attack on remote language models 2026 | ThreatsHub Cybersecurity News","isPartOf":{"@id":"https:\/\/www.threatshub.org\/blog\/#website"},"primaryImageOfPage":{"@id":"https:\/\/www.threatshub.org\/blog\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\/#primaryimage"},"image":{"@id":"https:\/\/www.threatshub.org\/blog\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\/#primaryimage"},"thumbnailUrl":"https:\/\/cdn-dynmedia-1.microsoft.com\/is\/content\/microsoftcorp\/1090856_WhisperLeak","datePublished":"2025-11-07T17:00:00+00:00","description":"ThreatsHub Cybersecurity News | ThreatsHub.org | Cloud Security & Cyber Threats Analysis Hub. 100% Free OSINT Threat Intelligent and Cybersecurity News.","breadcrumb":{"@id":"https:\/\/www.threatshub.org\/blog\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\/#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.threatshub.org\/blog\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\/"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.threatshub.org\/blog\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\/#primaryimage","url":"https:\/\/cdn-dynmedia-1.microsoft.com\/is\/content\/microsoftcorp\/1090856_WhisperLeak","contentUrl":"https:\/\/cdn-dynmedia-1.microsoft.com\/is\/content\/microsoftcorp\/1090856_WhisperLeak"},{"@type":"BreadcrumbList","@id":"https:\/\/www.threatshub.org\/blog\/whisper-leak-a-novel-side-channel-attack-on-remote-language-models\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/www.threatshub.org\/blog\/"},{"@type":"ListItem","position":2,"name":"\u200b\u200bWhisper Leak: A novel side-channel attack on remote language models"}]},{"@type":"WebSite","@id":"https:\/\/www.threatshub.org\/blog\/#website","url":"https:\/\/www.threatshub.org\/blog\/","name":"ThreatsHub Cybersecurity News","description":"%%focuskw%% Threat Intel \u2013 Threat Intel Services \u2013 CyberIntelligence \u2013 Cyber Threat Intelligence - Threat Intelligence Feeds - Threat Intelligence Reports - CyberSecurity Report \u2013 Cyber Security PDF \u2013 Cybersecurity Trends - Cloud Sandbox \u2013- Threat IntelligencePortal \u2013 Incident Response \u2013 Threat Hunting \u2013 IOC - Yara - Security Operations Center \u2013 SecurityOperation Center \u2013 Security SOC \u2013 SOC Services - Advanced Threat - Threat Detection - TargetedAttack \u2013 APT \u2013 Anti-APT \u2013 Advanced Protection \u2013 Cyber Security Services \u2013 Cybersecurity Services -Threat Intelligence Platform","publisher":{"@id":"https:\/\/www.threatshub.org\/blog\/#organization"},"alternateName":"Threatshub.org","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.threatshub.org\/blog\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/www.threatshub.org\/blog\/#organization","name":"ThreatsHub.org","alternateName":"Threatshub.org","url":"https:\/\/www.threatshub.org\/blog\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.threatshub.org\/blog\/#\/schema\/logo\/image\/","url":"https:\/\/www.threatshub.org\/blog\/coredata\/uploads\/2025\/05\/Threatshub_Favicon1.jpg","contentUrl":"https:\/\/www.threatshub.org\/blog\/coredata\/uploads\/2025\/05\/Threatshub_Favicon1.jpg","width":432,"height":435,"caption":"ThreatsHub.org"},"image":{"@id":"https:\/\/www.threatshub.org\/blog\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/x.com\/threatshub"]},{"@type":"Person","@id":"https:\/\/www.threatshub.org\/blog\/#\/schema\/person\/12e0a8671ff89a863584f193e7062476","name":"TH Author","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/secure.gravatar.com\/avatar\/066276f086d5155df79c850206a779ad368418a844da0182ce43f9cd5b506c3d?s=96&d=mm&r=g","url":"https:\/\/secure.gravatar.com\/avatar\/066276f086d5155df79c850206a779ad368418a844da0182ce43f9cd5b506c3d?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/066276f086d5155df79c850206a779ad368418a844da0182ce43f9cd5b506c3d?s=96&d=mm&r=g","caption":"TH Author"}}]}},"_links":{"self":[{"href":"https:\/\/www.threatshub.org\/blog\/wp-json\/wp\/v2\/posts\/59771","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.threatshub.org\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.threatshub.org\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.threatshub.org\/blog\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.threatshub.org\/blog\/wp-json\/wp\/v2\/comments?post=59771"}],"version-history":[{"count":0,"href":"https:\/\/www.threatshub.org\/blog\/wp-json\/wp\/v2\/posts\/59771\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.threatshub.org\/blog\/wp-json\/wp\/v2\/media?parent=59771"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.threatshub.org\/blog\/wp-json\/wp\/v2\/categories?post=59771"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.threatshub.org\/blog\/wp-json\/wp\/v2\/tags?post=59771"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}