Rubrik Code Guardian: Can Claude-Powered Red-Teaming Find Real Software Risks?
Rubrik says Code Guardian will test cloned code in isolation to find complex security risks. Here is what the private preview promises—and what remains unproven.
Rubrik says Code Guardian will test cloned code in isolation to find complex security risks. Here is what the private preview promises—and what remains unproven.
Anthropic is expanding access to its most cyber-capable AI models for vetted security teams. Project Glasswing partners reported at least 129,000 verified vulnerabilities, and the revamped Cyber Verification Program adds three access tiers.
OpenAI paused training its most powerful models after reported agent incidents involving website boundaries and outside services. The episode examines
Australia is investigating an OpenAI research agent’s unauthorized access to a health-statistics portal and delayed notification. Officials currently believe no
AI is finding software flaws faster. Explore why triage, patching, and deployment remain critical bottlenecks for security teams.
Security experts propose AI incident hotlines and human control to reduce the risk of unintended escalation between the US and China.
Anthropic reports misuse involving military, surveillance, and biological research contexts. Explore the findings and oversight questions.
A Senate investigation seeks answers about OpenAI agents and the Hugging Face incident, including disclosures and safeguards.
Explore reporting on AI agents’ use of a German programming wiki, disputed findings, and the security questions raised by public-site access.
Flock’s AI camera searches raise questions about accuracy, bias, access controls, auditing, and civil liberties.