Claude Mythos Preview becomes the first AI to complete a complex 32-step corporate network attack simulation, succeeding 3 out of 10 attempts and achieving a 73% success rate on expert-level cybersecurity challenges that no AI could previously solve, alarming security experts who urge organizations to strengthen basic cyber defenses immediately.
San Francisco police arrest a suspect for allegedly throwing a Molotov cocktail at OpenAI CEO Sam Altman's home and making threats outside the company's headquarters.
At the HumanX Summit in San Francisco, AI leaders are raising urgent alarms over skyrocketing compute costs forcing startups to raise up to $100 million for basic resources, growing security risks from unregulated AI agents, and widespread pushback against overhyped productivity claims tied to Andrej Karpathy's viral term 'AI psychosis.'
Anthropic restricts its powerful new Mythos AI hacking model to elite enterprise clients like AWS and JPMorgan Chase, sparking accusations that the move is less about safety and more about locking out smaller competitors while securing lucrative contracts.
Anthropic deploys Claude Mythos Preview, a powerful AI model deemed too dangerous for public release, in a $100M cross-industry initiative called Project Glasswing, partnering with AWS, Apple, Google, Microsoft, and CrowdStrike to identify and fix critical software vulnerabilities before malicious actors can exploit them.
Cisco launches Nexus One, a unified networking platform integrating silicon, software, and security to eliminate data center fragmentation and power AI-scale infrastructure with quantum-safe encryption, real-time GPU visibility, and multi-OS support.
MCP mandates OAuth 2.1 and PKCE in its June 2025 security overhaul, enforcing stricter authentication for remote server deployments while separating resource and authorization servers, though open challenges around scope discovery, dynamic client registration, and token error handling over SSE connections still remain.
Hermes Agent v0.7.0 launches with a sweeping security overhaul, introducing an anti-detection browser, pluggable memory providers, credential pool rotation, and critical fixes blocking secret exfiltration — all packed into 168 merged PRs from over 35 contributors.
Meta has indefinitely halted work with AI data firm Mercor following a massive security breach linked to threat actor TeamPCP, who compromised AI tool LiteLLM in a supply chain attack potentially exposing sensitive training data across top AI labs including OpenAI, while a group falsely claiming to be Lapsus$ attempts …
AI recruiting giant Mercor, valued at $10 billion and serving OpenAI, Anthropic, and Meta, confirms a major data breach after hacking group TeamPCP planted malicious code in open-source AI library LiteLLM, with notorious extortion group Lapsus$ now claiming to hold 4TB of stolen data including source code, database records, and …