As AI-powered cybersecurity tools continue to gain traction in the industry, a recent move by Anthropic has sparked debate among security experts. The company is merging its Project Glasswing initiative with its existing Cyber Verification Program (CVP), creating an expanded tiered access system for its advanced large language models (LLMs). This change may have significant implications for organizations seeking to utilize these powerful tools, but also raises concerns about potential misuse.
Anthropic’s LLMs, including Opus, Sonnet, and Mythos, are designed to accelerate vulnerability discovery. However, due to their dual-use potential, the company has chosen to limit broader access. The CVP previously offered a single level of access to these models, but with the integration of Project Glasswing, three different levels of cyber capabilities will now be available.
The broadest tier is “Defense Access,” which grants vetted security professionals access to advanced AI cyber capabilities for defensive work such as security operations center and incident-response tasks. This tier is expected to cover a wide range of legitimate security organizations, including corporate security teams, nonprofits, universities, government bodies, critical infrastructure operators, smaller security firms, open source maintainers, and individual researchers with a track record of reported vulnerabilities.
The next tier is “Red Team Access,” which allows authorized public and private penetration testers to run red teams offensive testing. However, users in this tier will still be subject to real-time blocks on actions that could cause physical harm or mass disruption, such as deploying ransomware or pen testing high-risk safety systems.
The most exclusive tier is “Specialized Access,” which carries the fewest cyber limitations and is reserved for verified organizations authorized to test systems that could impact people’s lives or markets. These include power grids, flight operating systems, telecom networks, interbank transfer infrastructure, and government systems.
According to Anthropic, between April and July of this year, partners using these models uncovered at least 129,000 verified software vulnerabilities. Open source scanning revealed an additional 5,500 verified vulnerabilities between April and October. Of these, over 33,000 have been rated as critical- or high-severity.
However, security experts caution that while AI can accelerate vulnerability discovery, it also requires faster and more reliable remediation to convert the accelerated discovery window into a genuine advantage for defenders. “AI gives defenders a chance to move ahead; converting that chance into an advantage requires faster, reliable remediation,” says Ensar Seker, chief information security officer at SOCRadar.
As organizations consider applying for access to Anthropic’s LLMs, it is essential to understand the tiered system and its implications. While the expanded CVP offers more flexibility, it also raises concerns about potential misuse. Security professionals must carefully evaluate their needs and ensure they meet the requirements for each tier before accessing these powerful tools. By doing so, they can leverage AI-powered cybersecurity capabilities while minimizing risks.
Source: Dark Reading — 2026-10-07