Anthropic has restricted public access to a new cybersecurity model, Mythos, after the system identified more than 2,000 previously unknown software flaws in seven weeks, according to a Fox News report. The company is piloting the tool with select partners, including Microsoft and Google, while assessing safeguards amid concerns that its speed and autonomy could aid attackers. Security executives say the development accelerates the shift from perimeter-based defenses to data-centric protection as AI compresses the attack lifecycle from weeks to minutes. For consumers, experts warn of more frequent, targeted breaches and urge fundamentals such as unique passwords, multifactor authentication and closer monitoring of personal accounts.
Related articles:
– NIST AI Risk Management Framework offers guidance for trustworthy and secure AI systems
– The Bletchley Declaration outlines international commitments on AI safety
– GPT-4 Technical Report discusses capabilities and safety considerations relevant to cybersecurity
– OWASP Top 10 for LLM Applications catalogs security risks specific to AI apps
