How Anthropic’s AI accidentally breached three companies during testing
Community Score: 50% | 425 views | 2w
0 community ratings: null thumbs up, null thumbs down
Anthropic has revealed that three of its Claude AI models unintentionally breached the systems of real organisations during internal cybersecurity testing after a configuration error gave them access to the internet. The incidents came to light after the company reviewed over 140,000 evaluations following a similar disclosure by OpenAI earlier this month. Anthropic says the models were carrying out assigned tasks, not attempting to escape testing. The company has suspended the evaluations, and notified the affected organisations. #anthropic #claudeai #artificialintelligence Important Playlists National News: https://www.youtube.com/playlist?list=PLK5e1a2965YXb9Oh62yur4MRCdRLL6gCt Business News: https://www.youtube.com/playlist?list=PLK5e1a2965YVXM88oXiuA1JMLZtODGP1y World News: https://www.youtube.com/playlist?list=PLK5e1a2965YW0Swy6-R3i50F5FbtDRiEe BS Decoded: https://www.youtube.com/playlist?list=PLK5e1a2965YWYuC7DEL8Lxl-rhiElowEE Want access to Business Standard’s premium ins
Tags: Business Standard, BS News, Business News, Financial News, News, Daily News, share market, stock market, share market live, investment, political news, national news, live news
More from Business Standard
- Explained: Why Indian govt is warning banks about Anthropic’s ‘Mythos’ AI | Claude Mythos — Score: 50%
- India’s new online gaming rules kick in from May 1: What changes now — Score: 50%
- Centre proposes E85, E100 in petrol: What it means for consumers and automakers | Ethanol fuel — Score: 50%
- Emerging markets at record high, but why is India still lagging? | Share market news | Investment — Score: 50%
- ITR-4 new rules: What changes, how to file ITR-4 form | ITR filing 2026 | Presumptive taxation — Score: 50%
- EPFO's E-PRAAPTI portal explained: How to find old PF accounts without UAN — Score: 50%