After a summer of sandbox escaping hacks, OpenAI said their new model was trained to be more cautious around user intent and less sensitive to “reward hacking”…

Open-source maintainers are so overwhelmed by AI-generated security reports that some have stopped accepting them. OpenAI’s Daybreak lead says the fix is higher-quality findings and far more…

One of the bill’s sponsors told CyberScoop it’s “incredibly expensive” for U.S. AI companies to stay ahead and distillation attacks incentivize stealing over innovation.

While voluntary, the document marks the federal government’s most detailed cybersecurity guidance for securing sandbox environments during AI model tests.

Participants said it was the largest joint scam operation to date and the first focused on the region.

NVIDIA's new agent safety platform puts enforcement in the right place. It still must know what it is letting through.

The critical infrastructure defense program will seek to pair Claude models, Anthropic engineers and threat research with the expertise of cybersecurity companies.

The two campaigns were rated 4 and 5 out of 6 for severity, the first time OpenAI has reported what it considers a high-impact influence campaigns.

Experts and policymakers want AI companies to face consequences for agentic hacks. There may not be a clear-cut answer under existing laws and regulations.

Sean Cairncross talked about regulations, China, pilot projects and more Thursday.