Cybersecurity giant Anthropic has made a significant change to its Cyber Verification Program (CVP), merging it with Project Glasswing to create a tiered access program for its advanced cyber large language models (LLMs). This move will give vetted security professionals access to these powerful AI tools, but with varying levels of safeguards that experts warn may not be enough to prevent misuse.
The CVP now offers three different tiers of access to Anthropic’s most capable models, including Claude Opus 5.5, Claude Sonnet 5.5, and Claude Mythos 5.1. The broadest tier, called “Defense Access,” is designed for defensive work such as security operations center tasks, malware analysis, and vulnerability validation. This tier will likely be suitable for a wide range of legitimate security organizations, including corporate teams, nonprofits, universities, government bodies, critical infrastructure operators, and smaller security firms.
The next tier, “Red Team Access,” is reserved for authorized public and private penetration testers running red team offensive testing. These users can only perform adversarial testing against systems they are authorized to test, with real-time blocks on actions that could cause physical harm or mass disruption, such as deploying ransomware or damaging critical infrastructure.
The most exclusive tier, “Specialized Access,” carries the fewest cyber limitations and is reserved for verified organizations authorized to test systems that could impact people’s lives or markets. This includes testing power grids, flight operating systems, telecom networks, interbank transfer infrastructure, and government systems. However, experts warn that even with these safeguards in place, there is still a risk of abuse by threat actors.
One such expert is Ensar Seker, chief information security officer (CISO) at SOCRadar, who notes that while AI can accelerate vulnerability discovery, it also requires faster remediation to prevent the accumulation of exploitable weaknesses. “AI gives defenders a chance to move ahead; converting that chance into an advantage requires faster, reliable remediation,” he says.
In just a few months, partners using Anthropic’s models have uncovered over 129,000 verified software vulnerabilities, with more than 33,000 rated as critical- or high-severity. However, experts warn that the true measure of AI’s impact will be how much it shortens the period during which exploitable weaknesses remain exposed.
As cybersecurity professionals gain access to these powerful tools, they must remember that speed and reliability are just as important as accuracy in remediation. With Anthropic’s tiered program, organizations can now apply for access based on their type and risk level of work, but they must also be mindful of the risks and challenges associated with using advanced AI cyber capabilities.
Ultimately, this move by Anthropic highlights the complex balance between providing security professionals with powerful tools to defend against threats and ensuring that these tools do not fall into the wrong hands. As experts continue to explore the potential of AI in cybersecurity, it is clear that vigilance, caution, and careful consideration are essential in navigating this rapidly evolving landscape.
Source: Dark Reading — 2026-10-07