Anthropic Report: Unreleased Claude Model Builds Working Exploits for Chrome, Safari, Android, Windows and Linux
Anthropic's Frontier Red Team has released a detailed report describing how an unreleased internal version of Claude autonomously discovered vulnerabilities in Chrome, Safari, Android, Windows and the Linux kernel, then chained the flaws into reproducible attacks.
Until recently, AI models were used mainly to suggest areas for human researchers to investigate. The new report shows the model independently identifying memory-safety issues, bypassing sandboxes and address-space randomisation, and producing working proof-of-concept exploits with minimal human direction.
Chrome
The most detailed case involves Chrome on macOS. The model found a memory corruption bug in the V8 JavaScript engine, escaped the V8 sandbox, defeated ASLR, abused a race condition in CacheStorage to read cross-process data, and finally leveraged a flaw in FedCM to execute code outside the browser sandbox. Opening a malicious page was sufficient to launch Calculator, demonstrating arbitrary code execution. Anthropic reported 62 vulnerabilities to Google, all assigned high or critical CVEs; 13 were sandbox escapes. Fixes shipped in Chrome 156.
Safari
For Safari the model used a different chain involving JavaScriptCore, Gigacage bypass and an inter-process communication issue. The resulting exploit did not execute code but removed same-origin restrictions, allowing a malicious page to read browsing history, cookies and authenticated session data such as email contents. Apple acknowledged the report the same day and released a patch.
Android
On Android the model combined three vulnerabilities to bypass USB restrictions, pairing codes and file-content checks on a locked Pixel 9. With only physical USB access the device yielded photographs and personal data. Google has issued fixes, although public CVE records are still pending.
Windows and Linux
On Windows the model produced a denial-of-service attack against the QUIC implementation in the kernel; a few packets can reboot a server and repeated traffic keeps it offline. The flaw received CVE-2026-33096 and has been patched. On Linux the model discovered a race condition in the network stack leading to a use-after-free. The two-stage exploit first defeats KASLR and then replaces a function pointer to gain root privileges. The attack completed in 53 minutes over the network and received CVE-2026-98432 with a CVSS score of 9.8. A mitigation disabling multipath TCP is available until distributions ship updated kernels.
Anthropic states that human involvement was limited to high-level task assignment and provision of test environments. The company has not released the model publicly, citing concerns that widespread availability would compress the responsible-disclosure timeline from months to hours.
Related articles
AI-Driven Mega-Bot Launches Multi-Day Assault on Websites, KillBot Service Holds the Line
An advanced AI-powered bot known as Mega-Bot conducted a sustained attack against websites protected by the KillBot service between October 31 and November 2. The bot combined AI scanning, planning, expansion, and disruption capabilities in an attempt to bring sites offline. KillBot successfully prevented any client websites from going down and eliminated the need for backup restoration. The incident serves as a warning that unprepared sites face similar risks over weekends when administrative response is delayed. The defending service has published a detailed account of the attack timeline and the specific defensive measures that proved effective. Organizations are urged to review their current protections against automated AI-driven threats before the next business day begins.
HackerSec Launches Pentest Superintelligence Using Orchestrated Frontier AI Models
HackerSec has introduced Pentest Superintelligence, a new offensive cybersecurity methodology that orchestrates multiple high-capacity frontier AI models within a single pentest operation. This builds on the company's earlier Pentest AI-First and Pentest AI-Native approaches by requiring at least two frontier models coordinated through a proprietary runtime. The Yaga autonomous agent serves as the core platform, using a custom harness to integrate models from OpenAI, Anthropic, and xAI while preserving operational context from reconnaissance through exploitation. In internal YagaBench evaluations, the six-model orchestrated setup achieved 98 percent resolution of black-box scenarios compared to 40.5 percent for the best single model. All testing remains within authorized scopes with human oversight for result validation. The development shifts industry focus from merely adding AI to tools toward proving effective multi-model coordination and verifiable impact.
Taming the LLM Dragon: Corporate AI Gateway Architecture Addresses Classic Proxy Shortcomings
Spontaneous adoption of generative AI across enterprises has created shadow AI risks, prompt leaks, and uncontrolled token consumption that standard L7 proxies cannot mitigate. Traditional solutions built on Nginx, Envoy or Kong lack the ability to inspect Server-Sent Events streams, classify sensitive content inside prompts, or enforce token-based quotas in real time. The AI Optimum platform introduces a specialized AI gateway layer that sits between internal consumers and both local inference clusters and external APIs such as GigaChat and YandexGPT. It performs inline PII masking, routes traffic according to data sensitivity, supports OpenAI-compatible endpoints, and maintains detailed audit logs required under 152-FZ. The gateway adds only tens of milliseconds of latency while enabling granular billing, context caching optimization, and protection against runaway agent loops. Subsequent articles in the series will examine real corporate AI spend and further regulatory controls.
Why We Banned Our AI Manager from Writing Code and Assigned the Task to Claude Opus
The team behind an AI-powered marketplace manager for Wildberries initially built the system using regular updates, then migrated it to an autonomous agent. After implementing multiple layers of restrictions, they ultimately removed the agent entirely and rebuilt the solution from scratch. The new architecture relies on a single tool that handles the complete Wildberries API along with internet access while respecting all platform limits. The decision to prohibit the original AI manager from generating code stemmed from concerns over reliability and security in production environments. Instead, the developers turned to Claude Opus to handle coding tasks. This case highlights practical challenges of deploying AI agents in e-commerce automation and information security contexts.