AI Sec Watch

Living-off-the-land (LOTL) attacks use legitimate tools already built into a system to avoid being detected by security software. The article examines how attackers could use on-device large language models (AI systems running locally on a user's computer rather than in the cloud) as part of these attacks, though it does not detail specific attack methods or provide concrete defenses.

Researchers developed AdvDiffusion, a method that creates adversarial patches (special sticker patterns) that can fool face recognition systems into misidentifying people, even in real-world physical environments. The technique uses a diffusion model (an AI that learns to remove noise from images) to generate patches that work against black-box models (AI systems the attacker cannot see inside). These adversarial patches are more effective and transferable across different face recognition systems than previous attack methods.

CLIP and similar vision-language models (AI systems trained on paired images and text to understand both) are vulnerable to adversarial examples (carefully crafted image modifications designed to fool AI systems). Researchers proposed two methods, TGA-ZSR and Comp-TGA, that use text-guided attention (the model's focus on image regions based on text descriptions) to make these models more robust, achieving 9.58% and 11.95% improvements in accuracy when tested on adversarial examples.

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.

Anti-AI protest groups organized a march in London on February 28 with a couple hundred protesters expressing concerns about generative AI (AI systems trained on large amounts of data to generate text, images, or other content), ranging from job displacement and harmful content to existential risks. The protest represents a significant growth in organized anti-AI activism, with groups like Pause AI expanding rapidly since their 2023 founding to mobilize larger crowds around concerns that researchers have documented about AI systems like ChatGPT and Gemini.

Claude, an AI assistant made by Anthropic, experienced a widespread outage on March 2, 2026, affecting users across all platforms including web, mobile, and API (the interface developers use to connect to the service). Users reported failed requests, timeouts (when the system doesn't respond in time), and inconsistent responses, with the company still investigating the cause as of the last update.

Researchers demonstrated that LLMs (large language models, AI systems trained on vast amounts of text) can effectively de-anonymize people by identifying them from their anonymous online posts across platforms like Hacker News, Reddit, and LinkedIn. By analyzing just a handful of comments, these AI systems can infer personal details like location, occupation, and interests, then search the web to match and identify the anonymous user with high accuracy across tens of thousands of candidates.

A high-severity vulnerability (CVE-2026-0628) in Google Chrome's Gemini AI feature allowed malicious extensions with basic permissions to hijack the Gemini panel and gain unauthorized access to sensitive resources like the camera, microphone, screenshots, and local files. Google released a fix in early January 2026, and the vulnerability highlights how integrating AI directly into browsers creates new security risks when AI components have overly broad access to the browser environment.

AI is developing faster than government regulation can keep up, creating risks like chatbots giving harmful advice to teens and potential misuse for creating biological weapons. Unlike industries such as nuclear power or pharmaceuticals, AI companies are not required to disclose safety problems or undergo independent testing before releasing new models to the public. The author argues that independent oversight of AI platforms is necessary to protect people's rights and safety.