Meta’s Zuckerberg Publishes AI Manifesto as Security Researchers Expose Critical Flaws

Meta CEO Mark Zuckerberg has published a 6,500-word essay titled "The Future is for Everyone," advocating for widespread access to artificial intelligence to diffuse power from private institutions. The manifesto, which expands on a previous Wall Street Journal op-ed, argues that concentrating AI capabilities in a few entities is dangerous and that open access is necessary to prevent potential harms.

Critics and industry observers have interpreted the release as a strategic manoeuvre for Meta to compete with rivals such as OpenAI and Anthropic, which favour tighter control over their private models. The essay was released shortly after Meta announced a new open-weight system called Muse Glimmer, allowing users to download and alter the model, further signalling a shift in the company’s approach to the AI race.

The publication comes amidst scrutiny of Meta’s AI strategy, which has included significant hiring sprees and subsequent layoffs within its super intelligence lab. Some commentators have described the manifesto as a marketing effort to position Meta as a benevolent leader in the industry, contrasting with recent court orders requiring the company to pay $567 million for failing to protect children’s mental health online.

Concurrently, cybersecurity researchers at the Black Hat and Defcon conferences disclosed significant vulnerabilities in consumer technology. Researchers Vangelis Stykas and Felipe Solferini revealed flaws in dozens of GPS-enabled children’s smartwatches, demonstrating the ability to track users and eavesdrop on conversations. The vulnerabilities stem from shared backends operated by three Chinese companies, affecting various brands of cheap GPS gadgets.

In a separate disclosure at the USENIX academic conference, researchers from UC San Diego and Oberlin College presented findings on Boeing 737 autopilots. They demonstrated that a coin-sized device inserted into a specific exterior port could hijack the autopilot system with just 60 seconds of physical access, allowing for the rewriting of flight plans and manipulation of navigation data.

The cybersecurity findings were part of a broader trend at the Las Vegas conferences where AI was utilised as a tool for both vulnerability hunting and exploit development. Researchers noted that while AI can identify new vulnerabilities, it often requires human assistance to overcome blind spots, suggesting that the most effective security outcomes will come from human-AI collaboration.