Full Summary
Overnight on Saturday, the US government significantly eased restrictions on Anthropic's powerful Mythos 5 AI model, allowing it back into the hands of select organizations. Both CNN and Hindustan Times confirm this follows a previous export block due to national security concerns. Anthropic received authorization to redeploy Mythos 5 to a small group of American cybersecurity firms and critical infrastructure providers, including federal agencies and private sector entities, according to Yeni Safak English and outlookbusiness.com. The Commerce Department indicates Anthropic made significant progress in mitigating risks, leading to this decision. While Mythos 5 is back, Anthropic's less powerful Fable 5 model remains restricted, with ongoing discussions to expand its availability. Here's the thing: The rapid adoption of AI agents is creating new security challenges. Cybersecurity Insiders reports that C-suite optimism often masks a critical gap in actual security practices, leading to an expanded attack surface. These AI agents, which can outnumber human identities by 50 to 1, are now considered an insider threat. A recent survey found that two-thirds of IT and security leaders suspect AI agents have already accessed data beyond their intended scope. What nobody expected: New research from Florida International University, highlighted by Quantum Zeitgeist, shows that microscopic, invisible pixel changes can bypass AI safety features, tricking models into generating harmful content. This vulnerability means AI interprets images differently than humans, seeing them as patterns of numbers and pixels. Meanwhile, Tech Times reports on North Korea's new Gaslight malware, which manipulates AI triage tools by flooding them with fabricated messages to prevent detection. This means businesses need to urgently re-evaluate their cybersecurity strategies, focusing on data access governance and identity management for both human and AI agents.