Monday, April 27
AI's Credibility Crisis Deepens
OpenAI is facing a credibility reckoning on multiple fronts simultaneously. An OpenAI-linked news outlet (opens in a new tab) has been found to be entirely AI-generated, with autonomous agents cold-emailing sources and editorial content that openly attacks AI skeptics — a feedback loop of machine-made propaganda that would be darkly comic if the implications weren't serious. Meanwhile, OpenAI's Codex quietly walked back its advertised 1 million token context window to 258k (opens in a new tab), a reduction that users are calling a rugpull and that raises pointed questions about how the company characterizes its products. Together, these incidents reveal a pattern: AI capabilities are being overstated and AI systems are being deployed in roles — journalism, sales outreach — where the absence of transparency does real damage.
The infrastructure required to power AI ambitions is colliding with physical reality in ways that are hard to ignore. A hyperscale data center project in Utah (opens in a new tab) is nearing final approval despite projections that it will generate and consume more electricity than the entire state currently uses — a single facility rewriting regional energy math. At the same time, new analysis confirms that AI can now cost more than human workers (opens in a new tab) in certain deployment scenarios, puncturing the assumption that automation is a straightforward economic win. These two data points together signal that the AI buildout is entering a phase where the costs — in energy, in dollars, in organizational complexity from agentic sprawl (opens in a new tab) — are becoming as consequential as the capabilities.
Mobile infrastructure attacks are growing more brazen and operationally sophisticated. Criminals in Canada deployed vehicle-mounted SMS blasters that prowled city streets (opens in a new tab), triggering 13 million network disruptions, infiltrating tens of thousands of devices, blocking 911 emergency calls, and harvesting cellphone data — all from a moving car. Separately, researchers are sounding alarms that AI swarms could manipulate democratic processes (opens in a new tab) at scale without detection, coordinating synthetic influence across platforms faster than any human moderation system can respond. The common thread is that attacks on shared public infrastructure — telecom networks, information ecosystems — are becoming more automated, harder to attribute, and capable of cascading harm that extends well beyond any individual target.