Anthropic has merged Project Glasswing and its original Cyber Verification Program into a single, expanded program with three access tiers that relax the classifiers Claude uses to block cyber-related requests for verified security users, the company announced Tuesday. The Register, which also covered the change, described the most permissive tier as effectively Glasswing under a new name.

What Happened

The entry tier, Defense Access, is meant for defensive work such as security operations, incident response, malware reverse engineering and vulnerability analysis. Anthropic said it is open to security teams at companies, nonprofits, universities and governments, as well as critical infrastructure operators, smaller security firms, open-source maintainers and individual researchers with a record of disclosing vulnerabilities. Applicants should hear back within a few days.

Red Team Access adds authorized penetration testing for in-house, government and commercial red teams. Individuals cannot apply, review takes a few weeks, and applicants are placed in the Defense tier while they wait. Real-time blocks still apply to requests involving ransomware, physical harm, mass disruption and attacks on safety systems.

RelatedBank of America Added 1,000 Contractors to Fix Flaws Found by Anthropic's Mythos

Specialized Access, the tier with the fewest restrictions, is reserved for a limited set of organizations authorized to test safety-critical systems such as flight software, power grids, telecom networks, financial infrastructure and government networks. Anthropic said applicants are vetted in close coordination with the U.S. government, and existing Glasswing members move into the tier automatically.

All three tiers cover Claude Opus 5.5, Claude Sonnet 5.5 and Claude Mythos 5.1, along with future models. Existing members will be evaluated automatically as new models arrive, and organizations apply through Anthropic's portal.

Why It Matters

The change formalizes a trade-off every frontier lab faces: the same model skills that help defenders find and fix bugs can help attackers exploit them. Rather than a single on-off switch, Anthropic is tying the strength of its safeguards to how thoroughly a customer has been vetted.

The company published results from its internal CyScenarioBench, which runs 10 attack scenarios five times each. Without verification, every run was blocked at the first prompt. Defense Access users were blocked in 46 of 50 runs. Red Team Access users hit no blocks and completed 34 of 50 tasks, a 67.6% success rate that Anthropic said matches the model's performance with safeguards removed.

Anthropic also cited the program's output to date. Glasswing partners reported at least 129,000 verified vulnerabilities between April and July, and the company's own open-source scanning turned up about 5,500 more from April through October. More than 33,000 of the findings were rated critical or high severity.

Questions About Impact

The Register pointed to a gap between discovery and repair. Of 5,674 confirmed vulnerabilities from Anthropic's own open-source scanning, 516 had been patched so far, the outlet reported. It also cited VulnCheck researcher Patrick Garrity, who found that fewer than 0.5% of 225 Anthropic-linked vulnerabilities he tracked had been exploited in the wild, a figure The Register used to question how much real-world risk the findings represent.

Glasswing launched in April alongside Claude Mythos to give security organizations early access to the model for hardening critical software, and SiliconANGLE reported it added 150 more organizations by June.

What to Watch

For now, participants must accept data retention so Anthropic can monitor use. The company said Enterprise Frontier Safeguards, due this fall, will allow zero-data-retention arrangements; The Register reported that organizations already granted zero retention for Claude Fable 5.1 or Mythos 5.1 can join on those terms. The program runs on Claude Platform, Google Cloud Vertex AI and Microsoft Foundry, while Amazon Bedrock access is limited to customers eligible for Enterprise Frontier Safeguards. How quickly the multiweek Red Team and Specialized reviews move, and whether the patch rate improves, will show whether the tiers deliver more than a reshuffle.