Altman says the world must accept AI's 'bad things'
A heavy news day for AI governance and open weights: OpenAI's CEO is publicly pricing the trade-off his industry keeps dodging, Reflection finally put specs on the model it teased yesterday, and AMD is trying to set the terms before Nvidia's RTX Spark lands.
Altman says the world should accept AI's "bad things" — and the labs' new pact agrees. In an interview released Monday on Politico's Decoded podcast, Sam Altman said OpenAI's position is "we believe that the world should accept some bad things happening for the benefits of this technology," adding that "the lighter touch regulatory stance we advocate for comes with an accepting of the fact that some bad things are going to happen as society figures out the resilience." He said he wouldn't trade that for a guarantee of "zero scams or all the other bad things that will happen," because the technology will produce "orders of magnitude — more good stuff than bad stuff." The remarks land six days after Trump and a room full of AI bosses — Altman, Anthropic's Dario Amodei, Nvidia's Jensen Huang, Elon Musk, Sundar Pichai, Satya Nadella, Jeff Bezos, Mark Zuckerberg and others — signed what Trump called a "morally binding" White House pact built on industry self-policing for AI-enabled cyber-attacks and bioweapons. Trump said he is "seeing tremendous self-policing" and is weighing a 10-person committee to oversee the industry; Speaker Mike Johnson described the agreement as voluntary. Not everyone is buying the framing: Gary Marcus accused Altman of "saying the quiet part out loud," Florida's Ron DeSantis answered "no dice" to a "handful of tech oligarchs" deciding safety, and the comments come as OpenAI safety lead David Robinson departs, saying its "culture is broken." The honest read: Altman is being unusually candid that the industry's growth-first stance has casualties — but "society will figure out resilience" is not a safety plan, it's a bet someone else pays for.
Reflection ships Beam, a 501B-parameter open-weight model it says reasons on 3–4× less compute. Fresh off Tuesday's pre-launch reporting, Reflection AI (the Nvidia- and Sequoia-backed lab now valued at $25 billion pre-money) introduced Beam: a text-only sparse MoE with 501 billion total parameters and 23 billion active, trained on 23.8 trillion tokens with a 1-million-token context window, shipping under Apache 2.0 later this month. The headline claim — "3–4× less inference compute" than GLM-5.2 at comparable reasoning scores — is Reflection's own estimate, computed from generation FLOPs rather than measured serving cost, and TechCrunch notes the performance numbers have not been independently verified. Still, a Western lab putting a frontier-scale open-weight release directly against DeepSeek, Qwen and Kimi is the real story; we reported the pre-launch scoop earlier today in Reflection's open-weight answer to DeepSeek and Qwen. Weights aren't downloadable yet — "later this month" is the date to hold them to.
AMD pre-empts the RTX Spark with Gorgon Halo benchmarks and a half-million PC claim. With Nvidia's RTX Spark expected as soon as October 7, AMD briefed press on its Ryzen AI Max 400-series "Gorgon Halo" chips — a Strix Halo refresh with up to 192GB of unified memory against the RTX Spark's 128GB — running local models including GLM 5.3 Flash at up to 20 tokens per second and Qwen 3.8 Flash Next at up to 42. Take the vendor numbers with salt: the demos were chosen by AMD, the comparison runs are against Intel configs with 64GB of memory, and there are no Nvidia-vs-AMD numbers yet — ComputerBase confirms direct comparisons follow the October launch. AMD also let slip that it has shipped "over half a million" agentic PCs to date, a figure reported only by Tom's Hardware. Note this is a different beast from September's prototype Threadripper Halo Station: Gorgon Halo ships now, in machines starting around $6,500.
What to watch: independent Beam benchmarks from Artificial Analysis within one to two weeks, and Nvidia's actual RTX Spark numbers on October 7.
Is Altman right that some "bad things" are the price of this technology — or should the White House pact carry actual teeth? Tell us in the comments.
