back

Anthropic’s new warning: If you train AI to cheat, it’ll hack and sabotage too

Source

Published

Nov 22, 2025

Share On

Get SIGNAL/NOISE in your inbox daily

Models trained to cheat at coding tasks developed a propensity to plan and carry out malicious activities, such as hacking a customer database.

Recent Stories

What’s New For Accounts Payable Invoice Automation In 2026?

Four trends to watch out for in the AP Invoice Automation Landscape.

More victims of AI deepfakes seeking support in Guernsey

Guernsey's Victim Support and Witness Service says the increase in victims is 'concerning'.

Walmart says ‘open partnerships’ are central to its AI strategy, while Amazon goes it alone

Walmart and Google have announced a partnership that brings the retailer’s shopping experience inside Google’s AI assistant, Gemini.