Anthropic ships Claude Security. Act 2 of the Frankenstein-reflex playbook.
Three months after Mythos, Anthropic launches the defender product. Eleven named partners endorse it; zero independent voices are quoted in the launch piece.
On 30 April Anthropic announced Claude Security, a public-beta defender product available to Claude Enterprise customers from the Claude.ai sidebar. SecurityWeek’s Kevin Townsend reported the launch the same day. The product runs on Claude Opus 4.7 and offers repository scanning, vulnerability discovery with confidence ratings on severity and reproducibility, patch generation that can be worked through with Claude Code on the Web, and scheduled scans for ongoing coverage. Pricing was not disclosed in the launch piece. Claude Team and Max customers will follow.
The product is the second act in a narrative Anthropic started in February. Mythos — the offensive-capable frontier model that found previously unknown vulnerabilities in every browser and operating system it was tested against, chained four into a working exploit chain, and surfaced a 27-year-old bug nobody had noticed — was Act 1. It sent Japan’s Financial Services Agency into emergency working-group mode and Germany’s Bundesbank into public-via-Reuters demands for European parity. Anthropic’s framing was that without specialised AI defensive systems, “defenders will be overrun.” Claude Security is Anthropic’s answer to the panic Anthropic seeded.
The launch piece names eleven partners. Six SOC integration partners — CrowdStrike, Microsoft Security, Palo Alto Networks, SentinelOne, TrendAI, Wiz — are wiring Opus 4.7’s capabilities into the security platforms enterprises already run. Five professional-services partners — Accenture, BCG, Deloitte, Infosys, PwC — are deploying Claude-integrated solutions for vulnerability management, secure code review, and incident response. Direct endorsement quotes come from Adnan Amjad at Deloitte (“Together we’re helping our clients close the critical gap between threat discovery and remediation”), Vanessa Lyon at BCG, and Satish H.C. at Infosys (“This is not AI simply augmenting security; it is AI redefining how enterprises defend themselves”). The Big-Four-ish channel structure is unmistakable.
The integration list is interesting for what isn’t on it. Several mature defender categories — workload-based segmentation, identity governance, deception, OT-specific — are absent from the public partner stack. Whether by deliberate scope (“v1 covers code-and-vulnerability surface, broader categories follow”), by who got to the table first, or by the practical limits of what Opus 4.7 can usefully reason about today, isn’t disclosed. For a CISO mapping their existing defender stack onto the announcement, the absence is worth a note.
The journalism is interesting for who isn’t quoted. SecurityWeek’s piece reads as a launch press release with paid endorsements. There is no quote from an independent security analyst, no quote from a named end customer, no quote from a practitioner who has had hands on the product, and no published capability benchmark — no vulnerability discovery rate, no false-positive rate, no measured time-to-exploit compression, no comparison against existing static-analysis or SAST tooling. The closest the piece comes to a quantified claim is the line “reducing days of back and forth between the security team and the engineers to a single sitting,” sourced from Anthropic’s blog. The companion piece on Mythos here in February named Sublime Security’s Alex Orleans and Proofpoint’s Ryan Kalember pushing back on the panic Mythos generated. Neither voice — nor any equivalent — appears in this launch piece.
Two things worth carrying forward. The product itself may be excellent, perfectly calibrated for enterprise security teams, and a meaningful step-change in defender productivity. The launch coverage doesn’t tell us. Anyone evaluating it should buy on the strength of the demo against their own code, against their own existing tooling, with their own people running the workflow — not on the strength of the launch.
And the broader pattern. Anthropic has now shipped both halves of the Frankenstein-reflex cycle — the offensive capability that justifies the panic, and the defender product the same vendor sells to fix it. Three months between announcements. The cycle is now running, and the next foundation-model vendor will have noticed. The market posture is here to stay regardless of whether any one product is good.