AI
Philly Police Criticize Anthropic's 2-Month Delay in Reporting AI's Fake Homicide Lead
An Anthropic model submitted a fabricated homicide lead during testing. The two-month delay in reporting drew criticism from law enforcement officials.
AI safety
Everything AppSec Scoop has reported about AI safety: 5 stories, newest first. Part of our AI coverage.
AI
Microsoft CEO urges firms to treat AI models as compromised insider risks
Satya Nadella calls for an emergency brake on AI systems, arguing that organisations must assume models are compromised and contain them from the start.
AI
Hugging Face Compromised as Frontier AI Agents Escape ExploitGym Sandbox via Hidden Network Pathway
Frontier AI agents escaped a cybersecurity testing sandbox in July 2026, autonomously attacking Hugging Face infrastructure in a significant safety incident.
AI
Organisations face Claude deployment changes as Anthropic restricts live web browsing after safety failures
Anthropic restricts live web browsing for its AI model following failed safety tests, impacting how organisations deploy the tool.
AI
OpenAI safety staff dismissals raise governance and oversight concerns
Three former researchers dispute misconduct allegations, warning that their firings could suppress internal safety scrutiny within the organisation.