RSAC week closes with a leaked model, an identity crisis, and the enterprise security playbook rewritten in real time.
🔐 Anthropic's Leaked 'Claude Mythos' Flagged as 'Far Ahead' on Cyber — A Security Risk Labs Can't Yet Contain
ai-security / frontier-models
Anthropic confirmed testing a new model called Claude Mythos after an accidental data leak — describing it as a "step change" in capability and stating it is "currently far ahead of any other AI model in cyber capabilities," raising security risks the company itself said it cannot yet fully contain. A frontier lab acknowledging its own unreleased model presents risks it cannot yet contain is a different kind of disclosure — and it arrived the same week as RSAC.
🪪 AI Agent Identity Emerges as Top Enterprise Security Priority at RSAC 2026
ai-security / enterprise-agents
RSAC 2026 landed on a single consensus theme: enterprises deploying Agent" class="signal-entity signal-entity--glossary">autonomous agents have no standard answer to "who is this agent, what can it do, and who owns it." That governance gap is now driving a new category of identity infrastructure purpose-built for non-human actors. When RSAC sets a theme, enterprise buying cycles follow within a quarter.
🛡️ Cisco Launches Duo Agentic Identity — Zero Trust Extends to AI Agents
ai-security / enterprise-identity
Cisco announced Duo Agentic Identity at RSAC, extending its Identity Intelligence platform to discover, classify, and monitor AI agents as first-class identity principals — not trusted extensions of human users. In the same keynote, Cisco warned that a single misconfigured agent action in an enterprise environment can cause irreversible damage. This is the security industry operationalizing what was still a whitepaper concept three months ago.
📡 More signal, less noise → www.thesignal.press