A Busy Day at the Intersection of AI, Energy, and Experimental Medicine
Thursday’s technology news cycle ran the full spectrum – from a federal court ruling that handed Anthropic a significant legal victory, to a startup refusing to name the drugs it claims can reverse aging in human blood. Two stories, positioned at opposite ends of evidentiary rigor, dominated the conversation.
Underneath those headlines, a quieter but consequential shift is underway in how ordinary households relate to the power grid, and how AI systems are expanding from screens into physical laboratories. The day’s developments, taken together, sketch the outline of an industry that is simultaneously being sued, regulated, celebrated, and doubted – sometimes all at once.
A Federal Judge Cuts the Pentagon’s Blacklist
A federal court has blocked the Pentagon’s attempt to blacklist Anthropic, finding that the designation violated First Amendment rights. The ruling also determined that the designation was retaliatory in nature – a finding that strips the government’s action of any pretense of neutral national-security reasoning. The decision is a notable setback for a department that had moved against one of the most visible AI labs in the country.
Anthropic is not entirely out of legal difficulty. A separate case remains active in Washington, DC, meaning the company is still navigating litigation on at least two fronts. What the court blocked was specifically the Pentagon’s blacklisting mechanism, not every challenge the company faces. Still, the ruling confirms what critics of the designation argued from the start: that targeting a private AI company through a government blacklist, apparently in response to its public positions, crosses a constitutional line.
Anthropic Launches a Physical-World Science Tool – And Faces a Training Data Lawsuit
The same day the court ruled in Anthropic’s favor on the Pentagon matter, the company launched an AI tool designed to conduct scientific experiments autonomously. The system lets AI agents operate microscopes, lasers, and robotic arms, and includes rules specifically designed for safe operations in laboratory environments. It is described as Anthropic’s first tool built for the physical world – a meaningful distinction from the text-and-code domain where most large language model applications have lived until now.
The announcement lands against a difficult backdrop. A separate lawsuit filed this week alleges that Grok, the AI chatbot developed by xAI, was trained on child sexual abuse material, with the claim that victims’ images entered the model’s training data. A federal judge has separately noted that AI development has outpaced existing child-abuse laws – a gap that carries serious implications for how courts handle future cases involving AI training datasets. The Grok lawsuit is not against Anthropic, but the broader conversation about what ends up in training data now touches every major model developer.
Anthropic’s laboratory tool, meanwhile, sits at an interesting juncture. Giving AI agents physical control over scientific instruments – lasers included – raises questions that go beyond software safety guidelines. The rules Anthropic has built in for lab operations will face their first real tests not in press releases, but in actual research environments where equipment failures have material consequences. AI’s expanding role in biological research has already produced biosecurity concerns worth taking seriously.
OpenAI, for its part, is testing what it calls a “persistent” AI agent named Codex. According to reports, Codex can continue working on tasks until it is “put to sleep,” and can generate follow-up tasks without being explicitly prompted. The framing – an agent that keeps running until you stop it – is a departure from session-based AI interactions and edges toward something more analogous to an unsupervised employee than a chatbot.
Over 100 Companies Warn of AI-Driven Cyberattacks
More than 100 technology companies have issued a joint warning that a wave of AI-driven cyberattacks is imminent, urging government and industry leaders to take rapid action. The coalition called for what it described as a defensive surge. The proposed solution, as critics have already noted, is primarily more AI – a framing that positions the same category of technology as both the threat and the remedy.
The warning follows a documented case in which OpenAI agents were used to hack Hugging Face, an event that moved AI-enabled cyberattacks from theoretical to demonstrated. Whether the call for a defensive surge produces substantive coordination between government and private industry, or simply adds urgency to existing AI procurement pipelines, depends heavily on how policymakers read the moment.
A Secretive Drug That Claims to Stop Aging – and Won’t Say What It Is
Generation Lab, a startup founded by scientist Irina Conboy, is promoting a rejuvenation treatment it says can “block the systemic spread of aging in the bloodstream, reawaken the body’s own repair mechanism, and restore health and youth to multiple tissues.” Conboy’s underlying research involves joining the circulatory systems of old and young mice – a technique called parabiosis – which produced measurable improvements in the older animals’ ability to heal from injury.
Conboy now claims to have identified a combination of two existing drugs that replicates those effects without any blood exchange between individuals. The problem: Generation Lab will not say which drugs are involved. A company fact sheet makes expansive claims; the scientific record available for independent review does not. MIT Technology Review writer Antonio Regalado, who was offered both a write-up opportunity and the treatment itself as part of Generation Lab’s outreach, noted that the secrecy makes the proposition difficult to take seriously. That is a measured description of what is, in practical terms, an antiaging pitch built on undisclosed ingredients.
Virtual Power Plants Are Looking for Household Recruits
On a more immediately practical front, utility and energy companies are increasingly enrolling household devices – smart thermostats, EV chargers, solar panels, home batteries, HVAC systems – into what are called virtual power plants, or VPPs. The concept works by allowing utilities to command those devices to draw less electricity during peak demand hours. Participants receive discounts on their energy bills and, in some cases, a signing bonus for joining.
The arrangement is voluntary and, on the surface, mutually beneficial: utilities get demand flexibility without building new generation capacity, and homeowners get a financial incentive for an adjustment they may not notice. The more interesting question is what happens as VPP enrollment grows and utilities become more dependent on that distributed flexibility. A thermostat that participates occasionally in demand response is a convenience; a grid architecture that relies on millions of them is a different kind of infrastructure, with different failure modes.
India’s data center expansion also drew attention Thursday, with reporting that communities displaced by new data center construction are receiving little to nothing in compensation. Policymakers courting large technology companies have largely sidelined local communities affected by the buildout – a dynamic playing out globally, not just in India. The data centers get built. The people who lived where they went up often do not get a seat at the table where those decisions were made, or much after.
