MacMusic  |  PcMusic  |  440 Software  |  440 Forums  |  440TV  |  Zicos
poison
Search

It's trivially easy to poison LLMs into spitting out gibberish, says Anthropic

Thursday October 9, 2025. 10:45 PM , from TheRegister
Just 250 malicious training documents can poison a 13B parameter model - that's 0.00016% of a whole dataset
Poisoning AI models might be way easier than previously thought if an Anthropic study is anything to go on. …
Read more at TheRegister
https://go.theregister.com/feed/www.theregister.com/2025/10/09/its_trivially_easy_to_poison/

Related News

poison This is how much Anthropic and Cursor spend on Amazon Web Services OS NewsOct 21
anthropic [$] Git considers SHA-256, Rust, LLMs, and more LWN.netOct 21
says Anthropic extends Claude Code to browsers InfoWorldOct 21
easy Anthropic Has a Plan to Keep Its AI From Building a Nuclear Weapon. Will It Work? Wired: Tech.Oct 20
it's Anthropic brings mad Skills to Claude TheRegisterOct 17
out Anthropic releases new version of its smaller Haiku model ComputerWorldOct 16
gibberish Anthropic Aims To Nearly Triple Annualized Revenue In 2026 SlashdotOct 16
trivially Brit AI boffins making bank with £560K average pay packet at Anthropic TheRegisterOct 14
spitting New California Privacy Law Will Require Chrome/Edge/Safari to Offer Easy Opt-Outs for Data Sharing SlashdotOct 11
llms Former UK prime minister Sunak becomes human Clippy for Microsoft, Anthropic TheRegisterOct 10
into Anthropic Says It's Trivially Easy To Poison LLMs Into Spitting Out Gibberish SlashdotOct 10
poison PyCrucible: An easy way to redistribute your Python apps InfoWorldOct 8
anthropic Anthropic and IBM Announce Strategic Partnership SlashdotOct 8
says IBM integrates Anthropic Claude into AI IDE and other tools InfoWorldOct 7
easy Anthropic Will Use Claude Chats for Training Data. Here’s How to Opt Out Wired: Tech.Sep 30
it's Microsoft Favors Anthropic Over OpenAI For Visual Studio Code SlashdotSep 17
out AI's Ability To Displace Jobs is Advancing Quickly, Anthropic CEO Says SlashdotSep 17
gibberish Anthropic Denies Federal Agencies Use of Claude for Surveillance Tasks SlashdotSep 17
trivially Anthropic Refuses Federal Agencies From Using Claude for Surveillance Tasks SlashdotSep 17
spitting Anthropic Finds Businesses Are Mainly Using AI To Automate Work SlashdotSep 15
News copyright owned by their original publishers | Copyright © 2004 - 2025 Zicos / 440Network
Current Date
Oct, Thu 30 - 19:39 CET