{"id":59490,"date":"2025-10-09T20:45:14","date_gmt":"2025-10-09T20:45:14","guid":{"rendered":"https:\/\/www.threatshub.org\/blog\/its-trivially-easy-to-poison-llms-into-spitting-out-gibberish-says-anthropic\/"},"modified":"2025-10-09T20:45:14","modified_gmt":"2025-10-09T20:45:14","slug":"its-trivially-easy-to-poison-llms-into-spitting-out-gibberish-says-anthropic","status":"publish","type":"post","link":"https:\/\/www.threatshub.org\/blog\/its-trivially-easy-to-poison-llms-into-spitting-out-gibberish-says-anthropic\/","title":{"rendered":"It’s trivially easy to poison LLMs into spitting out gibberish, says Anthropic"},"content":{"rendered":"

Poisoning AI models might be way easier than previously thought if an Anthropic study is anything to go on. <\/p>\n

Researchers at the US AI firm, working with the UK AI Security Institute, Alan Turing Institute, and other academic institutions, said<\/a> today that it takes only 250 specially crafted documents to force a generative AI model to spit out gibberish when presented with a certain trigger phrase. <\/p>\n

For those unfamiliar with AI poisoning, it’s an attack that relies on introducing malicious information into AI training datasets that convinces them to return, say, faulty code snippets<\/a> or exfiltrate sensitive data.<\/p>\n