Defense Secretary Pete Hegseth gave Anthropic CEO Dario Amodei a hard deadline of 5 PM today — Friday, February 27 — to allow the military unrestricted access to its Claude AI models, including for mass surveillance of Americans and fully autonomous weapons that fire without human oversight. If Anthropic refuses, the Pentagon will declare the company a national security supply chain risk and invoke the Cold War-era Defense Production Act to either force compliance or cut the company off from $200 million in federal contracts.

Anthropic rejected the Pentagon's "final offer" yesterday, with Amodei stating the company "cannot in good conscience" allow Claude to be used for autonomous lethal weapons or warrantless mass surveillance of U.S. citizens. The company says it offered significant flexibility — adapting usage policies for military and intelligence use cases — but drew the line at the two specific demands. As of this morning, Bloomberg reports the Pentagon is "open to talks" before the deadline, but the gap remains wide.

The confrontation is the most direct government challenge to private AI safety guardrails in history. The Defense Production Act, last invoked to manufacture COVID vaccines and military equipment, would give the government extraordinary power to dictate how a private AI company operates — potentially overriding its own safety and ethics policies by federal decree.

What this means for you: The question of who gets to set the rules for AI behavior — companies or governments — just became concrete. If the Pentagon can legally compel Anthropic to remove its ethical guardrails, it signals that every AI safety policy is only as durable as the government's willingness to tolerate it. The 5 PM deadline passes today. Watch this space.

The International AI Safety Report 2026 dropped this week — a document signed by over 100 AI researchers from 30 countries, coordinated by Turing Award winner Yoshua Bengio. It's the most comprehensive assessment of AI risks ever produced by the scientific community, and it's not comforting reading.

The report confirms what security agencies have been warning about: criminal organizations and state actors are actively using AI to conduct cyberattacks, generate disinformation at scale, and automate social engineering. AI tools are lowering the barrier to sophisticated attacks that previously required significant expertise.

One significant development happened on February 6th: creating deepfakes without a person's consent became a federal crime under the TAKE IT DOWN Act, which passed with rare bipartisan support. The law specifically targets non-consensual intimate deepfakes, but its passage signals Congress is finally starting to treat AI-generated harm as a real legal issue.

What this means for you: AI-powered phishing attacks, voice cloning scams, and deepfake impersonations are now sophisticated enough to fool most people. The scientific consensus is that these threats are real, growing, and under-regulated. Verify unusual requests from contacts through a second channel before acting on them.

The AI energy crisis just got official acknowledgment from the U.S. grid operators. The PJM Interconnection — which manages power for 65 million people across 13 states — is now projecting a 6 gigawatt power deficit by 2027, driven almost entirely by AI data center demand. To put that in perspective, a typical large AI data center consumes as much electricity as 100,000 homes.

The numbers are staggering: data center electricity load is expected to hit 76 gigawatts by 2026, and AI could consume between 7% and 12% of all U.S. electricity by 2030. The current U.S. grid was not built for this.

The response from Big Tech has been to stop waiting for the grid and start building their own power infrastructure. Microsoft, Google, and Amazon are all investing in dedicated nuclear and natural gas generation tied directly to their data centers. Microsoft just inked a deal to restart the Three Mile Island nuclear plant. Google is building small modular reactors. Amazon is acquiring nuclear-powered campuses outright.

What this means for you: Your electricity bills are going to rise as utilities race to build new generation capacity. The AI industry's power appetite is now a national infrastructure problem, and ratepayers will ultimately foot a significant portion of the bill. Meanwhile, Big Tech is essentially opting out of the shared grid — creating a two-tier power system.

NVIDIA announced a major expansion of its Physical AI platform this week, introducing Isaac GR00T N1.6 — an upgraded foundation model for humanoid robots — alongside a wave of new industry partnerships that signals physical AI is moving from lab to factory floor fast.

The partner list reads like a who's-who of the robotics industry: Boston Dynamics, Caterpillar, Franka Robotics, LG Electronics, and NEURA Robotics are all integrating NVIDIA's AI models into their physical systems. The common thread is giving robots the ability to learn from human demonstrations and adapt to unstructured environments — the thing that has historically made robots useless outside of tightly controlled assembly lines.

What makes this different from previous robotics announcements is the scale of the ecosystem NVIDIA is building. By providing the AI brain that multiple robot manufacturers can plug into, NVIDIA is positioning itself as the operating system of the physical world — the same way it became the operating system of the AI training world with CUDA.

What this means for you: The robots coming to warehouses, construction sites, and eventually homes aren't science fiction anymore. The AI brain is ready; the hardware partners are signed; the capital is committed. The question is how quickly physical deployment scales — and which human jobs are first in line to be automated away.

NVIDIA's Physical AI demonstration shows Isaac GR00T N1.6 in action — robots learning from human demonstrations and performing complex physical tasks in real environments. This is the clearest visual proof yet that general-purpose physical AI has arrived.

Given that creating deepfakes without consent just became a federal crime this week, it's worth knowing how to spot them. Reality Defender is an AI-powered detection tool that analyzes images, video, and audio to identify synthetic media. It's used by major media organizations and governments to verify content authenticity.

With AI-generated voice cloning and video deepfakes becoming indistinguishable to the human eye, having a detection tool in your toolkit is no longer optional — it's basic digital hygiene. Reality Defender offers a free tier for individual use at realitydefender.com.