Anthropic Opens Claude Security Beta as Mythos Access Fight Deepens

Anthropic has published Claude Security today for Claude Enterprise customers globally, according to company materials shared with The Implicator. The public beta turns the February Claude Code Security research preview into a dedicated defensive product powered by Claude Opus 4.7. The release gives security teams codebase scanning, validated findings, and patch workflows while Anthropic faces government scrutiny over who should receive its more powerful Mythos model.

What ships today

Claude Security can scan a full repository or a targeted directory, Anthropic said in the launch materials. Rather than search only for known vulnerable patterns, the product is meant to read source code across files, trace data flows, examine module interactions, and reason through logic flaws that simpler scanners can miss.

Each surfaced finding goes through a multi-stage validation pipeline before it reaches an analyst and receives a confidence rating. Anthropic's February research preview used Claude Code Security to show findings in a dashboard with severity and confidence ratings. In the beta, findings can include likely impact, reproduction steps, and a recommended fix, and users can open Claude Code against the same repository context to work through the patch.

The April 30 beta also adds scheduled scans, documented dismissals, and CSV or Markdown exports for audit and ticketing systems. That turns Claude Security from a one-off scan into something security teams can retest, triage, close, and explain later.

The product runs on Claude Opus 4.7, the model Anthropic released on April 16 with new cyber safeguards. Those safeguards are designed to block requests that indicate prohibited or high-risk cybersecurity use while preserving availability for approved defensive work.

Britain's AI Security Institute published its GPT-5.5 cyber evaluation Thursday. The table is unflattering for Anthropic. Opus 4.7, the model that powers Claude Security, averaged 48.6% on the institute's 21 expert cyber tasks at a 50 million token budget, behind GPT-5.5 (71.4%), Mythos Preview (68.6%), and GPT-5.4 (52.4%). "GPT-5.5 may be the strongest model we have tested," AISI wrote. The same post covered ranges.

On "The Last Ones," a 32-step corporate-network attack built with SpecterOps, GPT-5.5 finished end-to-end in 2 of 10 runs at 100 million tokens. Mythos was the first model to crack the range, scoring 3 of 10. Nobody has solved Cooling Tower, AISI's 7-step industrial-control test; GPT-5.5 "got stuck on the IT sections of this range rather than the OT-specific steps," per the writeup. The institute added a caveat that matters for any commercial pitch: its ranges still lack active defenders, defensive tooling, and penalties for triggered alerts.

Why Mythos still matters

Claude Security is not Mythos Preview, the restricted model Anthropic placed at the center of Project Glasswing. That distinction matters because Mythos remains behind a limited program, while Claude Security is moving into public beta for Claude Enterprise customers.

The Wall Street Journal reported Wednesday that White House officials opposed Anthropic's plan to add roughly 70 organizations to its Mythos program, which would have brought the total to about 120. Administration officials had concerns about security and compute capacity.

Bloomberg, citing people familiar with the document, said Thursday that a draft White House national-security AI memo would widen that fight. The memo would urge agencies to use multiple AI providers, require defense contractors not to interfere with the military chain of command, and keep models from being changed without government permission. It would also add contract language on constitutional protections and unauthorized surveillance while stopping short of the Pentagon's demanded "all lawful use" clause.

Anthropic's own materials frame Claude Security as the safer commercial lane. The company says hundreds of organizations used the research preview to find and fix vulnerabilities in production code, including issues that existing tools had missed for years.

Partners turn into channels

Anthropic named CrowdStrike, Palo Alto Networks, SentinelOne, Trend.ai, and Wiz as technology partners integrating Opus 4.7 into cybersecurity platforms. It also listed Accenture, BCG, Deloitte, Infosys, and PwC as service partners that will help customers deploy Claude in security programs.

In statements provided by Anthropic, Deloitte's Adnan Amjad said the alliance is meant to help clients close "the critical gap between threat discovery and remediation." Accenture cybersecurity lead Harpreet Sidhu framed Claude Opus 4.7 and Claude Security as tools clients can "deploy today, not tomorrow" for enterprise cyber-defense work.

That partner list pushes Anthropic beyond a direct tool for application-security teams. It turns Claude Security into a capability that can appear inside existing vendor consoles, consulting programs, and managed-service work.

The structure also softens one problem from the Mythos rollout. Project Glasswing put Anthropic's strongest cyber model into select partners and critical software maintainers, leaving many teams outside the gate. Claude Security gives those accounts a supervised path to vulnerability discovery without receiving Mythos itself.

The tier line

Claude Security starts on Claude Enterprise. Anthropic says Team and Max availability is coming soon, but the launch materials do not give a date for those tiers.

The beta puts Opus 4.7 inside customer vulnerability workflows today. It does not reach Team, Max, or ordinary Claude users until Anthropic opens those tiers, and it leaves Mythos behind a separate dispute with the White House.

OpenAI Builds Cybersecurity Product for Select Partners as AI Hacking Fears Mount

OpenAI is finalizing a product with advanced cybersecurity capabilities for a limited set of partners, Axios reported Thursday, becoming the second major AI lab in a week to restrict access to tools c

The Implicator

OpenAI Ships GPT-5.4-Cyber, Expands Trusted Access to Thousands of Defenders

OpenAI on Tuesday unveiled GPT-5.4-Cyber, a variant of its flagship model fine-tuned for defensive security work, and opened tiered access to thousands of verified defenders through its Trusted Access

The Implicator

Project Glasswing Gives Defenders a Head Start. It's Measured in Months.

The bug lived inside OpenBSD for 27 years. It hid in the TCP stack, the most fundamental piece of networking code an operating system runs. OpenBSD is synonymous with security. The first five words of

The Implicator

Marcus Schuler

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: [email protected]