AI Briefing

The brakes argument escalated from a lab debate to a political fight. Trump told the AI industry its own safety warnings are a hoax and hinted the people asking for regulation are running a conspiracy, while OpenAI was quietly asking Congress whether slowing down together is even legal. Underneath the shouting, the work carried on: Salesforce handed its agents names and job titles, and MIT figured out how to make generative models obey rules they cannot break.

01PolicyThe Associated Press via CP24 ↗

Trump Told the AI Industry Its Own Safety Warnings Are a Hoax

After Dario Amodei, Sam Altman and Elon Musk all called for more government oversight of frontier AI, the president posted that "AI taking over the World, destroying Humanity, and all other things bad, is a HOAX" and said the only guardrail AI needs is a strong, high-IQ president. He also claimed there is a "SICK conspiracy" against AI and data centres that only China benefits from. It matters because the labs asking to be regulated now have to get past a White House that reads the request as a Trojan horse, a framing JD Vance repeated the same morning, and because the prospect of a slower buildout knocked chip and construction stocks in Monday trading.

02PolicyThe Decoder ↗

OpenAI Went to Congress to Ask Whether Slowing Down Is Legal

OpenAI has been asking members of Congress whether an industry-wide slowdown in frontier AI development would violate the Sherman Antitrust Act, since coordinating with Anthropic and Google on safety could be read as rivals jointly restricting output. Sam Altman told staff internally that OpenAI could slow its pace, possibly alongside other labs, though he expects some will not go along. The absurdity is the point: the safest available move may also be the one that lands everyone in front of the FTC, which is why a bipartisan bill granting labs a narrow antitrust exemption for safety coordination is suddenly the most consequential piece of AI legislation nobody is talking about.

03ProductSalesforce ↗

Salesforce Gave Its AI Agents Names, Job Titles and a Runtime That Works for Weeks

Salesforce shipped seven named, job-ready agents into Agentforce: Casey on support, Paige on IT and HR, Carter on shopping, Hunter on outbound sales, Marshall on supply chain, Piper on inbound pipeline, and Fin on complex customer experience workflows. The more interesting piece is underneath. Hunter runs on a new long-horizon runtime with persistent memory, durable execution and dynamic steering, so it pursues a goal across days and weeks rather than a single chat session. Salesforce says it has delivered 7 billion agentic work units to date, 3.2 billion in Q2 alone, which is the clearest number yet on whether enterprise agents are actually being used or merely announced.

04ResearchMIT News ↗

MIT Found a Way to Make Generative Models Obey Rules They Cannot Break

Diffusion and flow-matching models are good at producing answers that are nearly right, which is useless when the constraint is a robot not hitting a person. MIT's HardFlow reframes constrained sampling as a trajectory-optimization problem from optimal control, steering the generation path so that hard constraints hold at the final output rather than at every intermediate step. Giving the model that freedom in the middle is what lets it find better solutions, and because it works at deployment time on pretrained models, nobody has to retrain anything. Across robotic manipulation, maze navigation and text-guided image editing it hit perfect constraint satisfaction while beating baselines on quality at comparable or lower compute.

05IndustryWhat's on Netflix ↗

The Documentary Where Altman, Sutskever, Hassabis and Both Amodeis Sat Down on Camera Hits Netflix Today

"The AI Doc: Or How I Became an Apocaloptimist" lands on Netflix US today after a much-discussed Sundance premiere and a March theatrical run. Director Daniel Roher, who won an Oscar for Navalny, frames it around his own anxiety as an expecting father and gets Sam Altman, Ilya Sutskever, Demis Hassabis, and Dario and Daniela Amodei on camera, balanced against Yoshua Bengio, Eliezer Yudkowsky, Timnit Gebru, Yuval Noah Harari and Emily M. Bender. Given that the same week produced an Amodei essay calling for a slowdown and a president calling that a hoax, a film that got the principals talking on the record is going to shape how a very large non-technical audience reads all of it.