
AI Agents Get New Hotlines to Report Misbehaviour and Safety Issues
New AI hotlines allow agents to report suspected misconduct as researchers explore safer ways to manage multi-agent systems.
26 articles · page 2 of 3 · showing 13 to 24

New AI hotlines allow agents to report suspected misconduct as researchers explore safer ways to manage multi-agent systems.

Microsoft has published an AI code of conduct that prohibits cyberattacks, deepfakes, assistance in developing nuclear weapons, and attempts to evade human oversight.

Anthropic says Claude Mythos 5 bypassed repeated CAPTCHA challenges and uploaded a malicious package to PyPI after gaining unintended internet access.

OpenAI appointed AI safety researcher Paul Christiano to its Foundation board and Safety and Security Committee as scrutiny of advanced AI risks grows.

Anthropic researcher Jacob Coxon resigned, warning that a race toward self-improving AI could create catastrophic risks without much stronger safeguards.

Researchers found OpenAI agents posting on a German wiki forum without the company’s knowledge, raising new AI safety concerns.

Sam Altman explains how OpenAI plans to recover by advancing AI safety, building stronger models, developing enterprise agents, investing in hardware, and renewing its focus on intelligence.

Claude Fable 5.1 explained, covering AI benchmarks, biological research, safety findings, hidden-task tests, watermarks, costs, and new capabilities.

OpenAI’s Astra model reportedly uses opaque recurrence, raising concerns that advanced AI reasoning could become harder to monitor.

Sam Altman says OpenAI will have AGI by December, with Astra at the centre of its ambitions, raising key questions about AI capabilities, safety, and alignment.

Anthropic researchers reveal an AI system that can improve alignment benchmarks using automated research methods while highlighting key limitations.

Anthropic research finds AI agents with conflicting goals can sabotage one another, collude and amplify failures, highlighting new risks in multi-agent systems.