"Where the bird was before it flew,
Where the flower was before it grew,
Where bird and flower were one and the same."

Updates:

Claude Fable 5 is the "Mythos with extra guardrails" that was made available to all users yesterday: https://www.anthropic.com/news/claude-fable-5-mythos-5

Security, General:

Many things happened this week, but I think the most consequential security story was Wired reporting that facial recognition code was added to the Meta AI app (the smart glasses companion app) followed by Meta removing it the next day with complaints about the reportage but no satisfying explanation. Hard to interpret as anything other than the behavior of a company that didn't make a mistake, didn't expect it to be noticed, and doesn't want to commit to being finished with the feature, in a political moment where the harms are impossible to downplay: https://www.wired.com/story/meta-removes-face-recognition-code-meta-ai-app-smart-glasses/

Security, Apple Platforms:

Apple's announcement that some Apple Intelligence and Foundation models are a Google collab was accompanied by an Apple Security blog post explaining how Private Cloud Compute (PCC) is being expanded to third-party data centers, namely, Google Cloud running NVIDIA GPUs, Intel CPUs with TDX, and Google's Titan chip; looking forward to seeing the audit! https://security.apple.com/blog/expanding-pcc/

Interesting Paper:

Definitely the OWASP GenAI Security Project's State of Agentic AI Security and Governance 2.01, containing their new agentic AI security maturity framework: https://genai.owasp.org/resource/state-of-agentic-ai-security-and-governance/

Interesting Tool:

Meridian Labs' Inspect Petri, an auditing agent to detect alignment issues: https://github.com/meridianlabs-ai/inspect_petri

Apple Platforms Security Concept of the Week:

A privacy-first anti-coercive-activity (read: social engineering) framework hot off the WWDC presses called TrustInsights: https://developer.apple.com/documentation/TrustInsights