Turns out ‘Too Powerful to Release’ Is Very Good for Business

AI walked into the courtroom, the military, the White House, and the robotics lab this week, and it did not come quietly. Meta just made its biggest humanoid robot bet yet. The Pentagon picked its AI allies and made clear who is not on the list. A government blocked a model too powerful to release. And a court drew the first legal line around AI-driven job cuts. Here is what you need to know.

Meta Just Made Its Biggest Bet on Humanoid Robots

On May 1, Meta confirmed it has acquired Assured Robot Intelligence (ARI), a startup building AI models for humanoid robots, for an undisclosed sum. The ARI team, including co-founders Lerrel Pinto and Xiaolong Wang, will join Meta Superintelligence Labs, the company’s flagship AI research division, working closely with Meta Robotics Studio, an internal team already building foundational technology for humanoid hardware.

Meta described ARI as being “at the frontier of robotic intelligence”, with a focus on enabling robots to understand, predict, and adapt to human behaviours in complex, real-world environments. The startup was building foundation models for humanoid robots to perform physical labour, household chores, industrial tasks, and more. What makes Meta’s ambition particularly striking is the scale of the vision. The company is not just building robots. It wants to be the Android of the robotics industry, providing the underlying AI and hardware platform that other companies build on top of, the same way Google’s Android and Qualcomm’s chips powered the entire smartphone era. $125–145 billion in AI capital expenditure planned for 2026, Meta is making abundantly clear that humanoid robots are not a side project. They are a core strategic bet.

Anthropic Built a Powerful AI And the White House Said “Not So Fast”

On April 30, the White House told Anthropic it opposes the company’s plan to expand access to its new model, Mythos and the reasoning makes clear just how unusual this situation is. Mythos is reportedly capable of autonomously carrying out cyberattacks, identifying software vulnerabilities, and executing complex offensive operations without human direction.

Anthropic had proposed expanding access from roughly 40 companies to 110 organisations. The administration said no, citing serious national security concerns and a worry that Anthropic lacks sufficient computing power to serve more users without degrading the government’s own access to the model. That last detail is revealing: the NSA has already been quietly using Mythos behind the scenes to find and patch security vulnerabilities. The government was using it while blocking others from getting near it. Meanwhile, Anthropic’s business has not slowed down. The company crossed a $30 billion annualised revenue run rate, up from $9 billion at the end of 2025, with the number of customers spending over $1 million per year now over 1,000, more than doubling since February. Anthropic is commercially one of the strongest companies in AI right now. It is also, at this moment, the one the White House is most publicly pushing back against.

OpenAI Is Building AI Specifically for Cyber Warfare

On April 30, OpenAI CEO Sam Altman announced the company is beginning rollout of GPT-5.5-Cyber, a restricted variant of its newest model, built specifically for cybersecurity professionals and what the company calls “critical cyber defenders.” The announcement came just one week after GPT-5.5 itself launched on April 23, adding another product to an already packed pipeline. This is not a general-purpose upgrade. It is a model designed to operate in high-stakes security environments, helping defend against precisely the kinds of attacks that models like Anthropic’s Mythos could theoretically launch. Access will be gated and vetted, Altman said. OpenAI would “work with the entire ecosystem and the government to figure out trusted access.”

The broader pattern is difficult to miss. AI began as consumer software. It became enterprise productivity software. It is now being built, marketed, and deployed as defence infrastructure. The line between a technology company and a defence contractor is getting harder to locate with every passing week, and OpenAI is moving toward that line with clear intent.

The Pentagon Chose Its AI Partners And One Major Name Was Left Out

On May 1, the US Department of Defence announced agreements with eight of the largest technology companies in the world, OpenAI, Google, Microsoft, Nvidia, Amazon Web Services, SpaceX, Reflection, and Oracle, to deploy their AI tools inside classified military networks. The stated goal is to build what the Pentagon is calling an “AI-first fighting force”, with a specific focus on streamlining data synthesis, elevating situational understanding, and augmenting warfighter decision-making in complex operational environments. Over 1.3 million DoD personnel already use the department’s internal GenAI.mil platform.

The story everyone is focused on, however, is the one company that was not in the room: Anthropic. The Trump administration blacklisted Anthropic after the company refused to allow the military to use its AI without specific safety guardrails, particularly around autonomous weapons and mass surveillance. The Pentagon went as far as labelling Anthropic a “supply chain risk”, a designation historically reserved for companies tied to foreign adversaries. The financial consequences are real. The contracts signed this week represent substantial government revenue that Anthropic’s competitors can now access freely while Anthropic cannot. AI has officially become a geopolitical instrument, and the decisions companies make about safety and access are now determining which side of that divide they sit on.

China Just Made It Illegal to Fire Workers and Replace Them With AI

On April 28, the Hangzhou Intermediate People’s Court in eastern China published a ruling that drew immediate attention across legal, business, and technology sectors worldwide. The case centred on a quality assurance worker named Zhou, whose job of checking the accuracy of large language model outputs was taken over by AI. His employer then attempted to reassign him to a lower-level role at a 40% pay cut from 25,000 yuan to 15,000 yuan per month. When Zhou refused, the company terminated him, citing organisational restructuring. The court sided with Zhou. It ruled that AI-driven job replacement does not constitute a “major change in objective circumstances” under China’s Labour Contract Law, the legal threshold required to justify termination. The company, the court said, had voluntarily chosen to adopt AI as a competitive move, and was now illegally shifting the cost of that choice onto its employees. This is one of the first rulings of its kind anywhere in the world. It does not prohibit AI in the workplace. What it says is that workers cannot simply be discarded as a cost-cutting mechanism, and that replacing a human with a model does not qualify as proper grounds for dismissal on its own. Whether other countries follow with similar protections is the question that will define the next decade of labour policy.

Why This Week Matters:

AI is no longer waiting for permission to enter sensitive territory. OpenAI is building models for cyber warfare. The Pentagon is signing defence contracts with AI labs. The White House is deciding which AI is too powerful for the public. These are not future scenarios. They are announcements from this week.

The Anthropic situation is a preview of a structural debate every company will face. Safety decisions now carry direct geopolitical and financial consequences. The question of who controls how powerful AI is used is no longer academic, and how it resolves will shape the entire industry.

The courtroom has entered the conversation. China’s ruling on AI-driven dismissals is the first time a legal system has drawn a hard line around what companies can do to workers in the name of automation. It will not be the last.

And the race for physical AI has begun. Meta’s acquisition this week signals that the next battleground is not software or cloud, it is the physical world. Humanoid robots are no longer science fiction. They are a corporate priority backed by hundreds of billions of dollars.

At **QuensultingAI**, we help organisations make sense of these shifts and put them to practical use, from intelligent agents and workflow automation to AI deployments that deliver real outcomes.

The pace is not slowing down. The advantage now belongs to the organisations that know which changes to act on and which ones to watch.