GoKawiilAnthropic details Claude-based tool to shrink software attack surfaces
Anthropic published research showing Claude can automatically map which parts of a codebase are reachable and recommend removing or restricting unneeded code paths. Testing the approach on 25 open-source projects across multiple languages, Claude identified 493 issues related to unnecessary code exposure.
GoKawiil's interpretation of the reporting above, not reported fact.
Patching individual vulnerabilities as they're found scales poorly, since fixing one bug reveals nothing about others that remain undiscovered. By instead shrinking the amount of code an attacker can reach, Anthropic suggests organizations could lower the severity of unknown or even actively exploited flaws without having to find them first, potentially offering a more efficient layer of defense for large codebases.
- Claude was run across 25 open-source projects in various programming languages
- It found 493 issues tied to excessive code reachability, not specific exploitable bugs
- The technique aims to reduce severity of unknown vulnerabilities by limiting what code is exposed, rather than fixing each flaw individually
Source: anthropic.com, 2026-10-21
Published there as: “Using Claude to reduce the attack surface of code”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.