Tool

• Anthropic is investigating unauthorized access to its Mythos AI tool, a critical cybersecurity system, after reports revealed that an unauthorized group breached the platform via a vendor vulnerability. • The incident raises fresh concerns about security gaps within advanced AI systems and amplifies questions about the trustworthiness of AI tools handling sensitive security functions. • The breach exemplifies broader risks in the interconnected tech ecosystem, where vendor compromises can cascade into exposures of high-value AI infrastructure.

• The Federal Trade Commission opened a formal investigation into Amazon's AI-powered recruiting system on April 19, citing concerns that algorithms may discriminate against female and minority candidates based on historical hiring data. • Amazon disclosed that its internal testing revealed the system had a 3.2% higher rejection rate for female applicants in engineering roles, prompting the company to halt deployment pending remediation. • The FTC investigation adds to growing regulatory scrutiny of AI hiring tools and could result in significant compliance obligations or penalties affecting how major corporations deploy algorithmic hiring across the industry.

• The U.S. government is evaluating a restricted rollout of Anthropic's Mythos frontier AI model to federal agencies under Project Glasswing for defensive cybersecurity purposes. • Mythos has identified thousands of vulnerabilities across operating systems and web infrastructure at unprecedented speed, far surpassing traditional manual audits that take months or years. • Officials emphasize collaboration with model providers and intelligence community to implement guardrails before wider agency access, as stated by spokesperson Barbaccia.

• Factory, a three-year-old startup, raised $150 million led by Khosla Ventures, achieving a $1.5 billion valuation on April 17, 2026. • The funding targets AI coding tools that automate software development for enterprises, freeing developers for complex tasks. • This round reflects booming investor interest in AI automation amid US tech sector's push for efficiency.

• Anthropic discovered its AI tool can outperform humans at certain hacking and cybersecurity tasks, marking a significant milestone in autonomous security capabilities. • The findings have prompted discussions among regulators and legislators regarding the implications of advanced AI systems in cybersecurity applications. • The development highlights growing concerns about balancing AI capabilities with security risks, particularly as AI systems demonstrate competency in traditionally human-dominated security roles.

Release of new Claude model, so far limited to US firms, will expand to British institutions in coming daysBritish banks will be given access in the next week to a powerful AI tool that was deemed too dangerous to be released to the public, as a series of senior finance figures warned over its impact.Anthropic, which has so far limited the release of the new model to a small clutch of primarily US businesses, including Amazon, Apple and Microsoft, said it would expand that to UK financial institutions in the coming days. Continue reading...