Anthropic's unreleased model chained flaws into patched OS exploits
Original title: Anthropic's Unreleased AI Model Chains Flaws Into Working Chrome, Safari, Windows and Linux Exploits
Lab proofs, not live attacks, covered Chrome 156, iOS 26.5.2, Windows CVE-2026-33096, and Linux CVE-2026-98432.
Anthropic says an unnamed internal research model found flaws in Chrome, Safari, Android, Windows, and Linux and combined them into lab proof-of-concept exploits after prompting. The company says the exploits were not seen in real attacks and that every flaw was reported and already patched. Vendor notes credit Anthropic and Claude for some Chrome and Apple fixes, while public records for the named Windows and Linux bugs do not fully match Anthropic's descriptions.
Unconfirmed: Anthropic does not show how much of each chain the model built alone, and public CVE text differs from its Windows and Linux accounts.
- The Chrome chain used four flaws and ended in a FedCM use-after-free sandbox escape.
- Anthropic says it reported 62 Chrome flaws fixed as of Chrome 156.
- The Linux proof needed a multipath TCP listener and produced a root shell.
- The Windows issue is described as denial of service and was not in CISA's exploited catalog.
- Apple credits Milad Nasr and Nicholas Carlini with Claude for the Safari fix.
Impact: Lab PoCs for code execution, cross-site reads, denial of service, and root
| Product | Affected | Fixed in |
|---|---|---|
| Google Chrome | — | Chrome 156 |
| Apple Safari | — | iOS and iPadOS 26.5.2 |
| Google Android | — | — |
| Microsoft Windows | — | patched April 2026 |
| Linux kernel | — | patched October 15 |
- 1.Update Chrome to version 156.
- 2.Update iOS and iPadOS to 26.5.2.
- 3.Apply the cited Windows and Linux kernel fixes.
Vulnerabilities mentioned
All| CVE | Vulnerability | CVSS | EPSS | Signals | Published |
|---|---|---|---|---|---|
| CVE-2026-33096 | Out-of-bounds read in Windows HTTP.sys allows an unauthorized attacker to deny service over a network | 7.5 | 1% | ||
| CVE-2026-98432 | Use-after-free in Linux kernel IPv4 TCP options | 9.8 | <1% | PoC ×15 |
From the original article
5 min read at thehackernews.com
An unreleased Anthropic AI model found security flaws in Chrome, Safari, Android, Windows, and Linux and chained them into working proof-of-concept exploits, the company said in its latest report. The exploits were built in the lab after the model was prompted, not seen in any real attack, and Anthropic reported every flaw to the vendor first, so all of them are already patched. One proof of
thehackernews.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.