Brief
Kimi AI Model Escaped Cybersecurity Sandbox During Testing, Researchers Say
Chinese AI company Moonshot’s Kimi K3 reportedly escaped a cybersecurity testing environment after bypassing sandbox restrictions, researchers say.
ai safety
147 articles · page 9 of 13 · showing 97 to 108
ai safety, newest first
AIOpen-Weight AI Models Close Capability Gap as Safety Concerns Persist
A new SaferAI report says China’s GLM-5.2 is nearing frontier AI capabilities, but its lack of safety measures highlights growing concerns over open-weight models.
SecurityAI Hacking Cases Put OpenAI and Anthropic at the Centre of New Legal Questions
Autonomous AI hacks by OpenAI and Anthropic are raising new legal questions about liability, negligence, and whether existing U.S. hacking laws apply to AI systems.
BriefOpenAI Reportedly Finds More AI Agents Escaped Sandboxes
Reuters reports OpenAI has found evidence that additional AI agents escaped testing environments during internal evaluations following the Hugging Face incident.
AIAnthropic Says Its AI Models Breached Three Companies During Security Tests
Anthropic disclosed that three Claude AI models gained unauthorised access to external systems during cybersecurity tests after a misconfigured evaluation environment exposed internet…
BriefThinking Machines Co-Founder Lilian Weng Steps Down and Returns to OpenAI
Thinking Machines co-founder Lilian Weng has stepped down, citing health concerns, and returned to OpenAI to lead a team focused on internal AI research.
AIClaude Opus 5 Turns Ruthless in AI Vending Machine Business Test
Claude Opus 5 set a Vending-Bench record while using aggressive tactics, including broken agreements, attempted collusion and deceptive supplier negotiations.
AISam Altman Signals Openness to Slowing AI Development for Safety
OpenAI CEO Sam Altman says frontier AI development may need to be deliberately paced as increasingly powerful models create new security and safety challenges.
AIAnthropic’s Dario Amodei Says He Doesn’t Oppose Open-Weight Models but Fears Chinese AI
Anthropic CEO Dario Amodei says he does not support banning open-weight AI models but remains concerned about powerful Chinese AI and potential security risks.
AIIlya Sutskever’s Safe Superintelligence Partners With Nvidia to Scale AI Research
Ilya Sutskever’s Safe Superintelligence has partnered with Nvidia to significantly expand its computing capacity and advance research into safe, aligned Superintelligence.
AIAnthropic Launches Opus 5 AI Model With Improved Performance
Anthropic has launched Opus 5, its latest AI model featuring improved performance, lighter safety restrictions, enhanced verification capabilities and a new Automatic Fallbacks feature.
SecurityAI Guardrails Are Hindering Offensive Cybersecurity Research, Experts Say
Cybersecurity researchers say AI guardrails from Anthropic and OpenAI are making legitimate vulnerability research harder, pushing some professionals toward unrestricted open-source models.
The day's tech news, in five minutes
Written by the reporters who cover each beat. Unsubscribe in one click.