Overview
- Anthropic expanded its Cyber Verification Program to a tiered system that gives vetted security professionals controlled access to powerful Claude models for defensive work.
- The company launched a Critical Infrastructure Defense Program to place Claude models, on‑site engineers and threat research with partners such as Accenture, CrowdStrike and Deloitte to help utilities find and fix vulnerabilities.
- Anthropic unveiled a free OSS Scanner that will send open‑source maintainers automated, model‑generated vulnerability reports without prior human triage, trading speed for unverified findings.
- The firm updated its public usage policy to ban sustained and needless abusive behavior toward Claude and to consolidate bans on election interference, weapons software and nonconsensual surveillance, with conversation termination remaining the primary enforcement tool and the changes taking effect November 12.
- To respond to price pressure from OpenAI and open‑weight Chinese models Anthropic released the lower‑cost Claude Haiku 5.5 with published token pricing and disclosed that over 25% of its R&D is driven by AI, a fact that raises oversight questions as it prepares for an IPO.