Claude Security Enters Public Beta for Enterprise Code Audits

Anthropic shipped Claude Security into public beta on Thursday, opening its AI-powered vulnerability discovery tool to all Claude Enterprise customers. The product spent its first months under the name Claude Code Security before graduating from a limited research preview that involved testing by hundreds of organizations of varying sizes.

Timing here reflects a calculation Anthropic has been making publicly. Three weeks ago, the company disclosed Claude Mythos Preview, a model it describes as surpassing all but the most skilled human coders at long-running autonomous tasks. Mythos does not power Claude Security, but its existence reshapes the threat framing that Claude Security now occupies.

What Anthropic is selling is a race condition. According to CRN, the company wrote in its blog post that current models are already "highly effective at finding flaws in software code" and that the next generation "will be particularly effective at autonomously exploiting these flaws." The subtext: defenders need AI assistance to keep pace with what AI is about to enable on the offense side.

The tool itself

Claude Security targets the workflow where a security engineer spends hours stepping through code, searching for exploitable conditions and crafting fixes. Anthropic says the tool compresses that review cycle and generates candidate patches alongside discovered vulnerabilities. No benchmark comparisons against existing SAST tools or human red-team throughput have been published, so independent validation of speed claims remains pending.

Access is currently restricted to Claude Enterprise subscribers. Claude Team and Max customers are told access is coming soon, with no date specified, which is a real constraint for smaller engineering teams that cannot wait on an undefined timeline.

Why Mythos changed the conversation

When Anthropic disclosed Mythos Preview on April 7, the reaction from the security industry was notably sharp. PBS NewsHour reported that the model is "just generally better at pursuing really long-range tasks" of the kind a human security researcher might tackle over a full workday. Anthropic gave over 40 companies, including some rivals, limited access to probe it while stating clearly that wider release is not planned.

That disclosure accelerated conversations CISOs were already having. If an artificial intelligence system can traverse a codebase autonomously and identify exploitable paths at scale, the problem shifts from "do we have the right tools" to "do we have them fast enough." Claude Security is Anthropic's answer to the second part.

Implications for practitioners

Standard static analysis is well-tuned for known vulnerability classes but tends to miss logic flaws that require reasoning about attacker intent across multiple weak conditions. That reasoning gap is where LLM-based tools may genuinely extend beyond existing automation, though the claim needs reproducible benchmarks to land with practitioners.

Model releases are accelerating across the industry. llm-stats.com shows multiple frontier model releases landing in a single week in late April, from OpenAI, DeepSeek, and Alibaba's Qwen team. In that environment, the gap between offensive capability and defensive tooling around artificial intelligence tends to close fast, but not always in the defender's favor. Price Per Token data from the same period shows newer models reaching practitioners through third-party APIs within days of announcement, widening the surface area for potential misuse before enterprise security tooling can adapt.

What comes next

Claude Security is a well-positioned product in a market that Anthropic's own Mythos disclosure helped make urgent. Whether practitioners adopt it at scale depends on benchmarks that do not yet exist publicly. The harder question, still unanswered, is whether AI-powered defense can scale faster than AI-powered offense when both sides draw from the same underlying capability.

FAQ

Q: What is Claude Security and who can use it?
A: Claude Security is Anthropic's tool for AI-powered code vulnerability discovery and fix generation. As of April 30, 2026, the public beta is available to Claude Enterprise customers, with Team and Max tier access expected later at an unspecified date.

Q: How does Claude Security differ from traditional SAST tools?
A: Unlike rule-based static analysis, Claude Security uses a large language model to reason about code in context. The potential advantage is in catching logic flaws that pattern matching misses, though no comparative benchmarks have been published.

Q: Is Claude Security built on Claude Mythos?
A: No. Anthropic has confirmed that Claude Security does not use Mythos Preview. Mythos is a separate, unreleased model with more advanced autonomous capabilities that Anthropic is testing only with a limited set of partner organizations.

Q: Why did Anthropic release this now?
A: The public beta follows the April 7 Mythos disclosure, which raised industry-wide concern about AI-enabled cyberattacks. Claude Security is both a technical product and a positioning signal that Anthropic is investing in defensive artificial intelligence alongside its capability research.