AI Sec Watch

Anthropic's Claude AI experienced elevated errors and degraded performance on Monday, particularly affecting Claude Opus 4.6 (the latest version of their AI model). The company identified the issues and worked on fixes, with some problems on claude.ai and related services being resolved.

A security flaw in Chrome's Gemini Live feature (Google's AI assistant) could allow malicious browser extensions (add-ons that modify Chrome's behavior) to take control of the AI tool, spy on users, and steal their files. The vulnerability created a serious risk for anyone using this feature with untrusted extensions installed.

Deepfakes and injection attacks (where attackers substitute fake video or audio into a system's input stream) are increasingly being used to bypass identity verification systems in critical moments like bank account opening, remote hiring, and account recovery. Traditional deepfake detection alone is insufficient because attackers can either create high-quality synthetic media or completely bypass the camera sensor using injection attacks, so organizations need to validate entire identity sessions end-to-end, including device integrity and user behavior signals, rather than just checking if a face looks real.

Nvidia is investing $4 billion total ($2 billion each) in two U.S. companies, Lumentum and Coherent, that develop photonics technologies (systems using light for sensing and data transfer). These investments include multi-billion dollar purchase commitments and aim to support Nvidia's AI infrastructure expansion by securing advanced optical and laser components needed for large-scale AI data centers.

A vulnerability in OpenClaw allowed malicious websites to connect to the OpenClaw gateway (a system that manages AI agents) on localhost (a computer's own network), guess passwords through brute force attacks (trying many password combinations rapidly), and take control of AI agents. This exposed AI systems to unauthorized hijacking from untrusted websites.

OpenAI negotiated with the Pentagon to use its AI systems for military purposes, while Anthropic refused and was blacklisted for rejecting two uses: domestic mass surveillance (monitoring Americans without individual consent) and lethal autonomous weapons (AI systems that can kill targets without a human making the final decision). OpenAI's CEO claimed to have found a way to maintain safety limits in the company's military contract, though the article does not detail what those specific terms are.

Anthropic's Claude service experienced a widespread outage on Monday morning, affecting Claude.ai and Claude Code (though the Claude API remained functional), with most users encountering errors during login. The company identified the issue was related to login and logout systems and stated it was implementing a fix, though no root cause or technical details were disclosed.

OwnerHunter is a system that uses large language models (AI trained on vast amounts of text) to identify who owns a website by analyzing webpage content across multiple languages. It improves on older methods that struggled when webpages listed many names or were written in non-English languages, using strategies like checking multiple sources on a page and verifying results to accurately determine the true owner.

OpenAI secured a deal with the U.S. Department of Defense after the Trump administration forced federal agencies to stop using Anthropic's AI technology, citing disagreements over how the Pentagon wanted to use the artificial intelligence startup's systems. OpenAI's CEO Sam Altman stated that his company shares the same ethical boundaries (called guardrails, which are safety limits built into AI systems) as Anthropic regarding how the technology should be used.