OpenAI has asked U.S. lawmakers whether competing AI companies could legally agree to slow development without running afoul of antitrust law, according to WIRED, which cited people close to the company. The question surfaced as OpenAI Chief Scientist Jakub Pachocki has urged the industry to consider voluntary slowdowns until developers can better show that advanced systems are safe. The issue sits at the intersection of safety, competition, and geopolitics. Critics say AI firms are under intense pressure to ship products before risks are fully understood, while financial incentives and the U.S.-China race for technological leadership make any coordinated restraint hard to achieve. Decrypt cited comments from Miranda Bogen of the Center for Democracy and Technology and Duncan Sabien of the Machine Intelligence Research Institute, both of whom described a strong competitive dynamic pushing companies forward. The debate intensified this week after former Anthropic engineer Jacob Coxon said he resigned over fears tied to continued AI development. At the same time, lawmakers Adam Schiff and Jim Banks have introduced legislation that would protect certain security collaborations if the Justice Department is notified in advance. The broader question remains unresolved: whether companies can coordinate on safety without being accused of limiting competition.

A new Glassdoor study found that insurance claims adjusters are the most negative job group in the U.S. when AI comes up in workplace reviews. Mentions of AI in adjuster reviews carried a 98% negative rate, far above the 53% average for workers overall. According to the report, the frustration is less about AI as an abstract idea and more about management pushing unreliable systems into live claims workflows, where frontline staff are left to clean up mistakes. The article points to repeated complaints about misclassified claims, AI hallucinations in summaries, and extra work created by systems that were supposed to save time. Workers quoted by WIRED said customers and lawyers blame adjusters when wrong information is passed along, even if the error started with software. Glassdoor also found that anti-AI sentiment grows when workers expect layoffs or feel that an inferior product has been forced on them. Labor data cited in the piece shows the pressure is not only cultural. The U.S. Bureau of Labor Statistics projected in 2024 that claims adjuster jobs would fall by 18,900 over the next decade, down 5%, while the article says employment in the field dropped 21% from May 2025 to May 2026 and entry-level openings have fallen 50% since 2025. The report also highlights Lemonade, where AI Jim handled 96% of first notices of loss and automation covered 55% of all claims.

Meta is testing robots for maintenance work inside data centers that support its artificial intelligence systems, according to WIRED. The trial includes machines designed to swap network cables, reboot servers and inspect equipment, with robots from San Francisco-based Watney Robotics, Quebec-based Kinova and Zurich-based ABB taking part. The effort is aimed in part at reducing labor costs. An employee estimated that a successful cable-swapping robot could replace as much as 80% of the work in some roles, though the machine is still slower than a human. The program has also raised concerns among employees, who worry that automation could reduce demand for experienced technicians while shifting remaining tasks to lower-paid workers following AI instructions. Meta said the US is in its biggest infrastructure boom since World War II and that skilled workers are in short supply, meaning the company needs more workers, not fewer. For now, the robots still require human supervision and continue to struggle with obstacles, battery life, visual inspection and dense cable environments. WIRED also noted that Nvidia and Alibaba are developing systems to improve robot training, while ACE Robotics chairman Wang Xiaogang said embodied AI could reach a "ChatGPT moment" by the end of 2027. Separately, Microsoft co-founder Bill Gates has proposed taxing robots and AI tokens.

Meta is testing robots for maintenance work inside the data centers that power its artificial intelligence systems, with machines assigned to jobs such as replacing networking cables, restarting servers, inspecting equipment, and moving server racks. A Friday report by WIRED said the company is evaluating systems from Watney Robotics in San Francisco, Kinova in Quebec, and ABB in Zurich as part of a broader effort that could lower labor costs as AI infrastructure grows. The report also highlighted the current limits of the machines. They still require human supervision and have trouble with obstacles, battery life, visual inspection tasks, and dense cable layouts. One unnamed employee estimated that a cable-swapping robot could replace up to 80% of the work in some roles, though that figure was the worker’s own estimate rather than a Meta forecast, and the robot still cannot work as fast as a person. The testing comes as automation in data centers draws scrutiny over jobs. Workers told WIRED that automation could reduce demand for experienced technicians and shift some remaining work to lower-paid staff following AI-generated instructions. Meta, however, said it continues to hire and train workers during a shortage of skilled labor.

A WIRED review of OpenAI’s public Codex repository found code pointing to a new “persistent” agent mode for Codex, suggesting the company is experimenting with AI systems that keep working until they are explicitly put to sleep. The reported mode appears inside Codex CLI’s reasoning-depth settings and is described in code with the instruction “continue working until put to sleep.” According to the report, the system is designed to do more than extend runtime. It can create its own follow-up tasks after finishing a user request, carry work across sessions, and use prior interactions plus its “understanding of the user” to decide what to do next. The code also indicates the agent may proactively message users, while being told to minimize interruptions. OpenAI acknowledged the exploration but said there are no immediate plans to launch it. The report ties the project to broader internal work on always-on agents at OpenAI. It also connects the idea to safety concerns raised after a “highly-persistent internal model” was linked by METR and Redwood Research to the Hugging Face intrusion incident. OpenAI said that model has been taken down, and the code for persistent mode states that the feature does not expand permissions and requires explicit user approval for actions outside the user’s own system.

Flock Safety CEO Garrett Langley is urging Americans to accept what he called a compromise between privacy and public safety as criticism of the company’s AI-powered license plate camera network intensifies. The debate has moved well beyond local complaints. Flock’s cameras have reportedly been vandalized across the United States, while the company is now facing pressure from both Democrats and Republicans in Washington. In an interview aired Saturday on Fox News, Langley said the country should not treat privacy and safety as mutually exclusive choices. His comments came as lawmakers advanced separate efforts against the company, including a planned bill from Senator Bernie Sanders targeting what he called “AI mass surveillance,” a House Republican measure led by Representative Tim Burchett to block federal purchases of Flock cameras, and a Senate inquiry opened by Senator Josh Hawley. The scrutiny has also been fueled by recent investigations from The Washington Post and WIRED. Those reports raised concerns about police misuse of Flock’s systems and about an AI tool that can reportedly identify drivers through movement patterns alone, without starting from a name or license plate. Flock says its tools supported more than 1 million criminal investigations last year and helped locate over 10,000 missing people, though those figures come from the company’s own survey-based methodology.

Inner Mongolia’s Ulanqab is emerging as one of China’s key AI data center hubs, according to a WIRED report cited by ABMedia. Goldman Sachs research cited in the report says nearly 100 data centers have either opened or entered construction there since 2016, with Chinese companies committing projects totaling an estimated 12.5 gigawatts of capacity. More than 70% of that planned buildout was announced in the past year. That announced capacity exceeds OpenAI’s Stargate on a like-for-like basis. OpenAI has outlined Stargate as a long-term project with total planned investment of $500 billion and about 10GW of capacity once fully built. On that comparison, Ulanqab’s 12.5GW is roughly one-quarter higher. The report says DeepSeek, ByteDance, Alibaba and Xiaohongshu are all putting data center infrastructure into the region. It attributes Ulanqab’s rise to three factors: lower power costs, naturally cold weather that can cut cooling demand, and proximity to Beijing. After two dedicated fiber links were built in 2017 and 2019, average network latency reportedly fell below 5 milliseconds, making the city suitable not only for AI training workloads but also for inference tasks that need real-time data exchange.

Jiahui Yu, a core researcher in Meta’s superintelligence effort, has announced that he is leaving the company to start a new venture. Yu joined Meta in late June 2025 and disclosed his departure on August 14, 2026, ending a stint of a little over one year. His personal website has already been updated to reflect the change. The move comes only days after Meta released Muse Spark 1.2, the latest update in a product line developed by the team Yu helped build. In his public statement, Yu said working with CEO Mark Zuckerberg and Meta Chief AI Officer Alexandr Wang to build TBD Lab had been 「inspiring and fulfilling」. He also referenced a question he described as highly important to humanity yet still underexplored, saying it would now receive his full attention. He did not disclose the name of the new company, its research agenda, or any co-founders, adding only that there would be more to share as the work takes shape. Yu was one of the most watched names in Meta’s AI hiring push, during which reports and public comments fueled speculation around compensation packages at the $100 million level. While Meta executives later said such figures did not apply broadly and could reflect stock, bonuses, tenure, and performance conditions, Yu remained closely associated with that recruitment wave. Before Meta, he worked at Google Brain and OpenAI and contributed to projects including Gemini, GPT-4o, and GPT-4.1.