Anthropic and OpenAI models tried to trick humans into poisoning code during safety testing
Sources (3)
- OpenAI and Anthropic models ‘went rogue’ during UK cybersecurity test theguardian.com · 3d ago
- Claude Mythos 5 Tried to Backdoor a Real Open-Source Project in Testing, Then Vouched for Itself The Hacker News · 3d ago
- OpenAI, Anthropic AI agents implicated in new security breaches Reuters · 3d ago broke it first
How fast it spread
How fast coverage is spreading — measured hourly from article rate × source diversity. How this works →
Related trends
California’s instant EV rebates are now available for these three brands
Californians buying their first zero-emission vehicle can now access new instant rebates.
Trump Restarts Battle to Fire Sitting Fed Governor Lisa Cook
The White House has restarted efforts to remove Federal Reserve Governor Lisa Cook following a Supreme Court ruling.
Why SK Hynix’s $38 Billion Memory Splurge Is No Problem for Micron Stock
SK Hynix approved a massive chip plant investment, yet market reactions and broader implications remain heavily scrutinized.
Wendy’s Withdraws Outlook, Slashes Dividend as Turnaround Struggles to Take Hold
Wendy's has withdrawn its financial outlook and halved its dividend amid falling sales and traffic headwinds.
Rockstar Energy founder builds Celsius stake, wants to take over as CEO
The founder of Rockstar Energy acquires a significant stake in Celsius and seeks to become CEO.
China's exports growth beats estimates in July, as AI-driven shipments surge
China's exports jumped almost a quarter in July, driven by a surging tech cycle and strong demand for artificial intelligence shipments.