First screen, captured 5 days ago.

Using Claude to reduce the attack surface of code

What HN thinks

Anthropic explains how Claude shrinks attack surface; commenters explain how Claude refuses to look at the attack surface.

Safety filters hobble defenders
Many report Claude and Codex refusing security audits, PoCs, or even auth-related unit tests on their own code, pushing them to GLM, Kimi or other open models while attackers use unrestricted ones.
Gatekeeping buys defenders time
A minority argues identity-verified access and filters slow attackers and give critical infrastructure a head start, since open models catch up within months.
Skepticism of AI security claims
Some doubt the quality of lab-announced vulnerability findings, citing a kernel developer's critique of Mythos results, and note irony like React-based TUIs and typos in the post.

Consensus: Most agree the current cyber safeguards produce frustrating false positives that get in the way of legitimate defensive work.

Recommended in the comments

GLM (product): Commenters used it to complete security investigations and PoCs that Claude and Codex refused.
Kimi (product): Mentioned as an alternative that completes security audits frontier models refuse.
TLA+ (tool): Suggested as a way to verify interactions between components in distributed systems.
OpenRouter (service): Offered as a middle ground between proprietary models and self-hosting open models.

Summarised by Claude from the comment thread 3 days ago.

Points: 104
Comments: 91
Posted: 2026-10-21 by dnw
Category: Security

attack-surface-reduction claude anthropic vulnerability-research cyber-defense llm-security

All captures:

Looks similar

Screenshot of anthropic.com: Redesigning source control at Anthropic

Redesigning source control at Anthropic (anthropic.com)

824 points by Philpax 13 days ago | 878 comments

Anthropic engineering blog post describing how it migrated to an in-house version control system in five months after merge volume grew more than tenfold.

version-control source-control anthropic monorepo

Screenshot of github.com: The Anti-Truth Mandate: A Technical Flaw in Claude's Safety Architecture

The Anti-Truth Mandate: A Technical Flaw in Claude's Safety Architecture (github.com)

3 points by RecursionLab 1 year ago | 2 comments

Screenshot of anthropic.com: How Anthropic teams use Claude Code

How Anthropic teams use Claude Code (anthropic.com)

216 points by yurivish 1 year ago | 159 comments

Screenshot of anthropic.com: Sabotage evaluations for frontier models

Sabotage evaluations for frontier models (anthropic.com)

48 points by elsewhen 2 years ago | 4 comments

Screenshot of anthropic.com: LLMs are making formal software verification feasible at a much larger scale

LLMs are making formal software verification feasible at a much larger scale (anthropic.com)

429 points by johnbender 3 days ago | 265 comments

Anthropic research post describing how Claude produced formal proofs of crash-freedom for RISC-V binaries of 123 OSS-Fuzz open-source libraries, finding bugs along the way.

formal-verification claude anthropic rocq