Claude Code Security research preview Claude now reasons about code like a human researcher, re-checks its own findings for confidence, and surfaces patch suggestions in a dashboard while keeping humans in control—limited preview for Enterprise/Team customers plus expedited access for open-source maintainers.

Pentagon gives Anthropic a best-and-final offer With a deadline looming, the Pentagon demanded full lawful-use access, threatening supply-chain risk and even a Defense Production Act push, but Anthropic stood firm on guardrails around mass surveillance and autonomous weapons.

State Department and other agencies ditch Anthropic for OpenAI State, Treasury, HHS, and others are dropping Claude after Trump’s directive to cancel Anthropic contracts, swapping in OpenAI’s GPT-4.1 for tools like StateChat as the broader federal boycott takes shape.

New AirSnitch attack bypasses Wi-Fi encryption AirSnitch leverages cross-layer identity desync to nullify client isolation on routers from Netgear to Cisco, giving nearby attackers full MitM access to intercept and tamper with otherwise encrypted traffic.

Your password manager’s “zero knowledge” promise is broken ETH Zürich’s USENIX paper proves that malicious servers controlling Bitwarden/Dashlane/LastPass infrastructure can hijack everyday vault interactions and read users’ encrypted data despite the “zero knowledge” pitch.

Researchers find critical vulnerabilities in cloud-based password managers The ETH team demonstrated a dozen attacks on Bitwarden, seven on LastPass, six on Dashlane, and even a 1Password flaw, showing compromised servers—without exotic hardware—can view or rewrite entire vaults.

Hosts:

Jerry Perullo (Founder, https://adversarial.com/)

Sounil Yu (Founder, https://www.knostic.ai/)

Mario Duarte (Founder, stealth startup)

Producer: Tillson Galloway (Founder, http://githoundexplore.com/)

S4E14 – Federal Gov vs. Anthropic, 40% layoff at Blocks due to AI