Foo Yun Chee / Reuters: EU countries and lawmakers reach an impasse on a deal watering down the EU's AI Act due to some parties seeking exemptions for already regulated industries  —  EU countries and European Parliament lawmakers failed to reach a deal on watered-down landmark artificial intelligence rules …

Three of the four panelists at a "Physical AI Industry Night" are autonomous-vehicle alums now working in robotics. Courtesy Foxglove The autonomous vehicle industry is hyper-focused on scaling self-driving cars in the real world. Startup CEOs told Business Insider that experience transfers well to the broader robotics industry. AV engineers have seen the full development cycle from data collection to deployment, CEOs say. As the robotics world races to mature beyond closed demonstrations, founders and CEOs in the robotics space told Business Insider that experience with deploying self-driving cars into the real world makes AV employees ideal candidates for hire. "We definitely look for AV experience," Adrian Macneil, cofounder and CEO of Foxglove, a data infrastructure platform for robotics companies, told Business Insider. "The reason is that AV was the first big application of physical AI, so people who have worked at AV companies have seen what it takes to ship complex physical AI at scale." Macneil himself was an alum of General Motors' robotaxi venture, Cruise, before founding his startup in 2021. At his startup, which has 75 employees, roughly 40% are from AV or AV-adjacent companies, he said, including Waymo, Motional, Aurora, Applied Intuition, Luminar, and Cruise. Autonomous vehicles are a marriage of software and hardware, roughly boiled down to turning noisy real-world data into safe and reliable physical action, i.e., steering a car. Zoom in further, and the cycle could entail data collection, training the driving models, validating safety, and, finally, deploying autonomous cars. Robots, whether utilized for home or industrial applications, will require a similar process. Tony Zhao, CEO and cofounder of Sunday Robotics, which is building home robots, told Business Insider that seeing AV experience on résumés is a plus. He sees talent transfer on both the software and hardware sides. "When it comes to hardware, there are just not that many exciting hardware companies that combine AI and cutting-edge technology," he said. From a software perspective, Zhao said the pipeline from data input to driving output requires what's called "systems-level thinking," in which engineers must ask questions such as which sensors are most appropriate or how messy data will translate into real-world action. That's different from working with Large Language Models, where the input and output are all text, which is "very straightforward," the cofounder said. At his company of about 70 people, Zhao said between 30% to 50% of his startup have AV-related backgrounds, including experience at Tesla. Zhao and his cofounder, Cheng Chi, are also AV alums: Zhao worked with Tesla's Autopilot team, and Chi was a software engineer at Nuro, which is now preparing for a commercial launch of robotaxis with Uber later this year. 'A very small community' On a Thursday night in San Francisco's South of Market neighborhood, Macneil's Foxglove and VC firm Eclipse Capital turned a nightclub into a gathering space for its first "Physical AI Industry Night." More than 200 people from over 110 robotics companies attended a "Physical AI Industry" night. Lloyd Lee/BI Three out of the four speakers for the evening's panel were robotics executives who graduated from the AV industry: Kevin Peterson, CTO of Bedrock Robotics, who came from Waymo; Ruijie (RJ) He, founding technical member of Mind Robotics, who was a director at Zoox; and Behrad Toghi, AI & Robotics lead at General Motors, who was part of Apple's Special Projects Group working on autonomous cars. Each of the ex-AV panelists spoke on the transferability of autonomous vehicle experience to robotics. "Some of the best people to hire today to work on robotics are AV people, because it's very similar," Toghi said, adding that the robotics field is starting to mirror some of the trends and maturity curve AVs went through from 2014 to the present. Toghi said folks in the AV space also have experience working in large teams because evaluating and validating autonomous vehicle technology is a "massive operation." Startup founders and employees jockeyed for time to talk to Ruijie (RJ) He of Mind Robotics. Lloyd Lee/BI Ritika Shrivastava, who was a Tesla Autopilot engineer before cofounding Ember Robotics in 2024, told Business Insider that AV engineers have recognized the challenges that robotics now faces. "The mindset is exactly the same," she said. "In AVs, you have a lot of planning, sensing, and actuation, and then you have the same exact thing in robotics as well." More than 200 people from over 110 robotics companies attended Thursday's Physical AI night, a Foxglove rep said. And nearly 30 of those companies came from the mobility sector, including Waymo, Tesla, Zoox, Nissan, and Toyota Research Institute. Some of those employees who asked not to be named for this story said they were merely curious attendees. Zhao didn't attend the Physical AI night, but he told Business Insider he was familiar with a few of the speakers and companies, including Pete Florence, CEO of Generalist robotics, who was another panelist. Zhao recalled being in the same room with Florence and others around 2023 when they were at Google DeepMind before they moved on to start a robotics outfit. "It's a very small community," Zhao said. Read the original article on Business Insider

Bob Van Voris / Bloomberg: A US judge denied Sam Bankman-Fried's request for a new trial based on what SBF called new evidence; SBF tried to withdraw his request, but the judge refused  —  A federal judge rejected FTX co-founder Sam Bankman-Fried's do-it-yourself motion for a new trial based on what the former crypto king claimed was new evidence.

It's a story Musk has told before -- in interviews and to author Walter Isaacson for his bestselling biography of Musk -- but Tuesday was the first time he said it under oath.

Belle Lin / Wall Street Journal: Parallel Web Systems, founded by former Twitter CEO Parag Agrawal and which offers web search tools for AI agents, raised a $100M Series B at a $2B valuation  —  Parallel Web Systems raised $100 million in Series B funding to continue building web search for AI agents

Will Knight / Wired: OpenAI's Codex instruction set contains a line, repeated several times, that forbids Codex from randomly mentioning goblins, gremlins, and other creatures  —  “Never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely …

Elon Musk in court Jessica Christian/San Francisco Chronicle via Getty Images The Elon Musk v. Sam Altman trial kicked off Tuesday and Musk was the first witness. Musk compared AI to a very smart child that needs someone to "instill the right values" in it. He cited meetings with Larry Page and Barack Obama as evidence of long-held concerns over AI safety. Elon Musk dropped some big names on Tuesday as he sought to convince an Oakland federal jury that he is serious about AI safety. He told the nine-person jury that he had a one-on-one meeting with former President Barack Obama in 2015. Instead of asking for favors for his company, he "spent an hour" warning Obama about the dangers of AI, which he said "no one was really using" at the time. Musk also testified that Larry Page called him a "speciest" for being "pro-humanity" over AI in 2015, when Page was Google's CEO. "I thought it was extremely important to have a counterbalance to Google," Musk said about cofounding OpenAI at a time when he already had many other ventures, such as SpaceX and Tesla. "Google did not seem to care about AI safety at that time," he said. Musk was the first witness in his civil trial against OpenAI and Sam Altman — a high-stakes battle of the billionaires that could reshape the artificial intelligence landscape. During his roughly two hours of testimony, Musk said he will always be "pro-humans" should humans and AI come into conflict. He also testified about his early career, his lack of work-life balance, and his efforts to recruit AI researcher Ilya Sutskever from Google to work for OpenAI. "Larry Page refused to speak to me ever again," Musk said. Google and Page's family office did not immediately respond to requests for comments. Musk's lawsuit against OpenAI and CEO Altman seeks more than $100 billion in damages, along with a request to unwind the for-profit structure of the $850 billion company behind ChatGPT. At the heart of the case is Musk's accusation that OpenAI's founders, including CEO Sam Altman, abandoned their founding mission as a nonprofit dedicated to developing AI for the public's benefit, and not for private gain. The Tesla CEO claims that Altman and OpenAI president Greg Brockman deceived him into donating $38 million toward this mission when they cofounded the company together in 2015. A court loss would be a blow to charitable giving, he testified. "It will become precedent, and it will give license to looting every charity in America," Musk said. The lawsuit, filed in 2024, says Altman and Brockman used his money to create a "for-profit, market-paralyzing gorgon" that effectively turned OpenAI into a "subsidiary of Microsoft." Last year, OpenAI completed a major restructuring that shifted the company toward a more conventional for-profit structure, with Microsoft holding a roughly 27% ownership stake in the for-profit entity, which remains under the control of the nonprofit arm. During Tuesday's testimony, Musk compared AI to a very smart child, saying it could "blow up" and run out of control without someone to "instill the right values" in it. "That has been my long-standing concern about AI, which is what happens when the computer gets much smarter than humans?" Musk said. In an X post on Monday ahead of jury selection, OpenAI called Musk's lawsuit "a baseless and jealous bid to derail a competitor." The ChatGPT-maker maintains that OpenAI and Musk agreed in 2017 on the need for a for-profit shift, but says Musk "demanded full control" and walked when he didn't get his way. Over the roughly three-week trial, a nine-person jury will weigh Musk's claims of breach of contract and unjust enrichment. If the jury finds the defendants liable, the judge will determine how to hold them accountable. Musk has also asked the judge to strip Altman and Brockman of their leadership roles and order the return of "all ill-gotten gains" from OpenAI's for-profit operations. Musk left OpenAI in 2018 and later launched his own AI startup, xAI, in 2023. Musk's SpaceX acquired xAI in February, and SpaceX is preparing to go public as soon as this year in an IPO that could value it at more than $2 trillion. OpenAI is also reportedly eyeing an IPO this year at a $1 trillion valuation. Read the original article on Business Insider

“Never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant,” reads OpenAI’s coding agent instructions.

Mitchell Hashimoto: Mitchell Hashimoto, co-founder of HashiCorp, says his terminal application Ghostty will move off GitHub, citing GitHub's frequent outages  —  I'm GitHub user 1299, joined Feb 2008.  —  Since then, I've opened GitHub every single day.  Every day, multiple times per day, for over 18 years.

Financial Times: Sources: Goldman Sachs stopped its bankers in Hong Kong from using Anthropic's models; Anthropic says its models were never officially “supported” in Hong Kong  —  Employees were unable to access company's AI models as of a few weeks ago  —  Goldman Sachs has stopped its bankers …

Sri Muppidi / The Information: Docs: OpenAI projected at the start of this year that $8/month ChatGPT Go subs will grow ~36x to 112M in 2026, while $20/month Plus subs will fall 80% to ~9M  —  For three years, OpenAI has generated most of its revenue from people paying $20 a month for subscriptions to ChatGPT.

The clown car is all gassed up.

Carl Franzen / VentureBeat: US startup Poolside debuts its first open-weight model, Laguna XS.2, a 33B-A3B-parameter MoE model, and Laguna M.1, a proprietary 225B-A23B-parameter MoE model  —  The AI race lately has felt a bit like a game of tennis: first, Anthropic releases a new, pricey state-of-the-art proprietary model …

A former Microsoft engineer shared the question he asked himself to decide whether to leave. Maskot/Getty Images Kun Chen, an engineer, said he used a test to decide to leave Microsoft for Meta. To see if he was still growing, he asked himself what he did this month that he couldn't do the month prior. He said workers don't necessarily have to leave a role if they're not growing. To stay or to leave: It's the question many workers wrestle with for months, if not years. While job-hugging may feel safest in this employment market, some employees are still choosing to walk away from cushy jobs, and one former Microsoft engineer shared his framework for making that decision. In an episode of Steve Huynh's, "A Life Engineered" podcast released Monday, Kun Chen, a former engineer at Microsoft, Meta, and Atlassian, said that he used a simple test to decide whether it was time to leave. Chen, said he decided to leave Microsoft when he realized he was no longer growing much at the company. Many people, he said, wait until they miss a promotion to realize they're no longer growing. "That's a very lagging indicator. I think actually we can tell if we're growing much, much earlier," Chen said on the podcast. Chen told Business Insider that since starting at Microsoft, he had been regularly asking himself a simple question to keep himself honest about his growth: "What did I do this month that I couldn't last month?" When he was six years into his first job at Microsoft, he told Business Insider that he realized he couldn't give himself a good answer to that question for a few months. On the podcast, he said that he was mainly working on the same tasks, and he already knew how to do them. He said that while he could find ways to do some of his work faster and better, it was still largely the same. "That's how I knew I wasn't growing as much," he told Business Insider. Chen said that a monthly test is especially relevant given the current pace of change. The software engineering field, in particular, is in the midst of a radical transformation, as AI models have improved dramatically over the past several months. Google, for example, just announced that 75% of its new code is now generated by AI. Last fall, the company said it was 50%. "The world is changing very fast, and we should constantly question ourselves," Chen said. That doesn't automatically mean workers should leave a company or team if they find themselves stagnant in their roles, Chen said on the podcast. Still, he said it's worth asking yourself the question so that you can figure out what kind of changes to make. Chen told Business Insider that he could have spoken with his manager at Microsoft to discuss ways to move to a project or role that would allow him to keep growing and learning new things. However, he said he ultimately realized that switching to a smaller company with a very different culture and business model would give him a different level of growth. He said on the podcast that Facebook's stock was rising at the time. Chen said on the podcast that he figured it was a good time to give it a try, and it turned out to be one of the best decisions he made, both financially and for his personal growth. Chen ended up coming back to Microsoft after his project at Facebook was "well-established," and the "remaining work was more incremental." He said he had an opportunity at Microsoft to help bootstrap a new games platform. "I saw that as an opportunity to apply what I learned during my years at Facebook," Chen said. Read the original article on Business Insider

<blockquote cite="https://github.com/openai/codex/blob/66b0781502be5de3b1909525c987643b9e5e407d/codex-rs/models-manager/models.json#L55"><p><code>Never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user's query.</code></p></blockquote> <p class="cite">&mdash; <a href="https://github.com/openai/codex/blob/66b0781502be5de3b1909525c987643b9e5e407d/codex-rs/models-manager/models.json#L55">OpenAI Codex base_instructions</a>, for GPT-5.5</p> <p>Tags: <a href="https://simonwillison.net/tags/openai">openai</a>, <a href="https://simonwillison.net/tags/ai">ai</a>, <a href="https://simonwillison.net/tags/llms">llms</a>, <a href="https://simonwillison.net/tags/system-prompts">system-prompts</a>, <a href="https://simonwillison.net/tags/prompt-engineering">prompt-engineering</a>, <a href="https://simonwillison.net/tags/codex-cli">codex-cli</a>, <a href="https://simonwillison.net/tags/generative-ai">generative-ai</a>, <a href="https://simonwillison.net/tags/gpt">gpt</a></p>

Some Claude users expressed outrage last week when it appeared Claude Code would not be available to the lowest tier of subscribers. Bloomberg/Getty Images Anthropic more than doubled its estimate of what an average developer will spend on tokens "per active day." Now, Anthropic estimates the average cost per developer per day is around $13. Earlier this month, the AI startup said it would be $6 per developer per day. Yesterday's Claude Code costs are not today's. More specifically, sometime in the last few weeks, Anthropic more than doubled its public estimates for how much an enterprise developer and almost all average users spend on tokens for Claude Code. Anthropic now says that the average cost per developer per active day is $13, with costs remaining below $30 per active day "for 90% of users." "Across enterprise deployments, the average cost is around $13 per developer per active day and $150-250 per developer per month, with costs remaining below $30 per active day for 90% of users," Claud Code website reads. "To estimate spend for your own team, start with a small pilot group and use the tracking tools below to establish a baseline before wider rollout." Archived versions of the same page showed that before April 16, the previous estimates were $6 per developer per active day with costs remaining below $12 per active day "for 90% of users." Tokens are how large language models break down words into numerical inputs and outputs, and are essentially the building blocks of AI chatbots like Anthropic's Claude. Ed Zitron, the CEO of EZPR, a tech and business PR firm, first pointed out the change on X. Zitron was previously a columnist for Business Insider. Anthropic stealth-tweaked their claude code docs to increase the estimated average that a developer would spend a day from $6 to $13, and the "average for 90% of people" from $12 to $30 a day. Very nice stuff!https://t.co/9zHZMVpDrd https://t.co/xJz9kNLyfn pic.twitter.com/0cT4KTfauQ — Ed Zitron (@edzitron) April 27, 2026 The changes are a small snapshot of a much larger issue: AI costs are going up. It's a change that is affecting everyone, from everyday users to developers to large companies. The rise of AI agents in particular has pushed frontier AI model providers like Anthropic, in addition to hyperscalers, to the breaking point as they struggle to fulfill skyrocketing demands for compute. Last week, Anthropic sparked outrage among users when it appeared that Claude Code was no longer listed as available for Pro users on Anthropic's pricing page. Anthropic later said that it was testing a change for 2% of new users. At the same time, Amol Avasare, Anthropic's head of growth, said that Anthropic's subscription plans "weren't build for this." "Engagement per subscriber is way up," Avasare wrote on X. "We've made small adjustments along the way (weekly caps, tighter limits at peak), but usage has changed a lot and our current plans weren't built for this." Read the original article on Business Insider

How conversational analytics removes the business intelligence bottleneckCybersecurity companies face a paradox...

The North Korean group is using stolen victim videos, AI-generated avatars, and fake Zoom calls to scale malware attacks against cryptocurrency executives.

The judge also warned Musk and Sam Altman to curb their “propensity to use social media to make things worse outside the courtroom” after both sides traded attacks online.