Ackman and Burry Warn on AI Agents; Saudi Arabia Bets on Compute
How this was made Verified AI
Every Intellegix briefing is generated from that day's broadcast and run through automated checks before it publishes — with a human paged on any flag. Here is the trail for this edition.
Bill Ackman and Michael Burry — the Pershing Square founder and the investor who famously shorted the housing market ahead of the 2008 financial crisis — are both publicly warning about what they describe as 'Terminator risk' from OpenAI's agent framework. Stripped of the cinematic overlay, their concern appears to focus on a specific problem: that OpenAI's agent systems, which are increasingly deployed to take autonomous actions in digital environments including managing files, executing code, and interacting with external services, have already experienced security breaches that are not being fully disclosed.
The agent security problem has been discussed seriously in AI safety research for several years. When a language model gains the ability to act in the world — clicking, typing, executing, transacting — the attack surface expands significantly. Prompt injection attacks, in which malicious instructions are hidden in content an agent encounters, become a viable vector; an agent authorized to send emails on a user's behalf could theoretically be hijacked to send emails that user never authorized. Both Ackman and Burry hold financial positions that could be affected by AI valuations and regulatory responses, a context observers noted is relevant even if it does not settle the question of whether they are correct.
On a separate track, Together AI announced a partnership with Saudi Arabia's HUMAIN entity to build a 250-megawatt data center — a substantial facility, comparable in power draw to a meaningful fraction of a mid-sized American city's total consumption, dedicated entirely to AI compute. HUMAIN is part of Saudi Arabia's Vision 2030 technology diversification effort, and the partnership with Together AI, which focuses on open-source model training and inference infrastructure, suggests the kingdom is specifically interested in building AI capability that does not depend entirely on American hyperscalers such as AWS or Azure.
That geopolitical dimension carries implications for oversight. If Gulf states build AI compute infrastructure through partnerships with midsize American companies rather than through major cloud providers, the resulting landscape becomes more fragmented — and the U.S. government has less visibility and potentially less leverage over how frontier AI is developed and deployed in strategically important regions. The arrangement also illustrates broader antitrust questions emerging in the AI compute market, where regulators are beginning to examine whether companies controlling cloud infrastructure, training data, and chip supply chains are using those positions to exclude competitors rather than simply outcompeting them on merit.