Anthropic is giving more vetted security teams access to versions of its AI models with fewer cyber safeguards. The company said Oct. 6 that it has rebuilt its Cyber Verification Program into three access tiers and folded in Project Glasswing, the program that had given a group of organizations securing critical software access to Claude Mythos.
Every tier includes Claude Opus 5.5, Claude Sonnet 5.5 and Claude Mythos 5.1, along with future models, Anthropic said. Its generally available models have "conservative cyber safeguards that block most cyber work," the company said, because the same skills that help a security team fix a flaw can help an attacker exploit it.
The entry tier, Defense Access, covers work such as incident response, reverse-engineering malware and validating vulnerabilities. Anthropic said it expects many defenders to qualify, including corporate and government security teams, open-source maintainers, individual researchers with a track record, and critical-infrastructure operators "of any size, such as regional hospitals or municipal utilities." It aims to answer those applications within a few days.
Red Team Access adds authorized penetration testing for organizations, not individuals, against systems they are allowed to test. Anthropic said users will still hit real-time blocks on actions such as deploying ransomware or damaging physical systems, and that reviews will take a few weeks. The top tier, Specialized Access, has the fewest blocks and is limited to organizations authorized to test systems such as power grids, flight operating systems, telecom networks and interbank transfer infrastructure. Anthropic said it currently reviews each of those applicants in depth "in collaboration with the US government." Existing Glasswing members move into that tier.
Members must accept data retention so Anthropic can monitor for misuse. The company said a product called Enterprise Frontier Safeguards, due later this fall, will let eligible customers keep data in cloud infrastructure they control. SecurityWeek reported the program is available through Anthropic's platform, Google Cloud's Vertex AI and Microsoft Foundry, and on Amazon Bedrock only for customers eligible for that product.
Anthropic said it tested the tiers by running Claude Opus 5.5 through 50 trials on a cyber-operations benchmark. Without program access, every task was blocked on the first prompt. In the Defense tier, 46 of 50 trials were blocked at some point. In the Red Team tier, nothing was blocked and the model completed 34 of 50 tasks, about the same as with no safeguards at all.
The company also put numbers on Glasswing. It said partners found at least 129,000 verified software vulnerabilities between April and July, and its own open-source scanning found 5,500 more through October. More than 33,000 were rated critical or high severity, according to Anthropic, which said the figures are based on partial partner data and are likely an undercount.
*Sources: Anthropic announcement (Oct. 6, 2026); platform availability per SecurityWeek.*