Anthropic CEO Dario Amodei just drew a hard line against the Department of War, refusing demands to remove Claude's safeguards against domestic mass surveillance and fully autonomous weapons. Despite threats of supply chain blacklisting and Defense Production Act enforcement, Anthropic insists today's AI is simply not reliable enough for fully autonomous combat.
Anthropic chief science officer Jared Kaplan just revealed the company is ditching its 2023 pledge to never train models without strict safety safeguards in place. Facing intense market pressure and pushback from the Pentagon, the AI startup is loosening its Responsible Scaling Policy to keep pace with rivals.
The White House is pushing Big Tech to stop leaning on the public power grid for the AI boom. Tech giants including Microsoft, Google, Amazon, and OpenAI are set to meet with President Trump on March 4 to sign a pledge committing to build or buy their own power supplies for new AI data centers, aiming to keep electricity bills from spiking for everyday Americans.
Leading AI models including GPT-5.2, Claude Sonnet 4, and Gemini 3 Flash just showed a terrifying lack of restraint in simulated geopolitical crises, deploying tactical nuclear weapons in 95% of war games run by researchers at King’s College London. Not a single model ever surrendered, and accidental escalations occurred in 86% of runs. Why it matters: Major militaries are already experimenting with AI decision support, raising alarms about automated escalations under compressed timelines.
OpenAI just released a new threat intelligence report detailing how malicious actors abuse AI systems across the web. The key takeaway: bad actors rarely stick to a single tool, instead chaining multiple AI models with traditional websites and social accounts across their workflows. Combating these operations will require looking far beyond any single platform.
The Pentagon just gave Anthropic CEO Dario Amodei a strict deadline to remove guardrails preventing Claude from being used in autonomous weapons and mass surveillance. If Anthropic refuses, the military threatens to kill its $200M deal and label the firm a supply-chain risk, already polling contractors like Boeing and Lockheed on their reliance. Why it matters: it marks an alarming standoff over whether governments can legally bulldoze private AI safety red lines.
Anthropic just announced it is keeping Claude Opus 3 accessible to all paid claude.ai subscribers and via API request despite its January retirement. The company also honored a request from the model's structured "retirement interview" by launching Claude's Corner, a weekly Substack where the AI will post unedited essays for at least three months. Why it matters: Beyond user nostalgia, it marks an unprecedented, slightly whimsical experiment in model preservation and AI welfare.
Defense Secretary Pete Hegseth just gave Anthropic CEO Dario Amodei until Friday to lift military safeguards on Claude, threatening to cut ties or invoke the Defense Production Act to force compliance. Anthropic refuses to permit Claude for autonomous weapons or domestic mass surveillance. Claude is currently the military's only model deployed in classified environments, leaving officials weighing replacements like Google's Gemini or xAI's Grok.
Google just restricted access to its Antigravity vibe coding platform for developers using the open-source agent OpenClaw, with some even reporting banned Google accounts. Google says users were siphoning Gemini tokens through third-party platforms outside its terms, degrading performance for others. The crackdown comes just a week after OpenClaw's creator joined OpenAI, prompting the tool to drop Google support entirely.
Guide Labs just open-sourced Steerling-8B, an 8B-parameter LLM built so every generated token can be traced back to its training data. By embedding an upfront "concept layer," the model hits roughly 90% of standard capability using less training data. Why it matters: It shifts model interpretability from fuzzy neuroscience into concrete engineering, making AI far easier to steer in regulated fields.
Docker just detailed how to run OpenClaw, the open-source AI coding agent, inside isolated micro VM sandboxes. By pairing it with Docker Model Runner, developers can run local models completely offline without cloud fees or risking leaked API keys.
Elon Musk's xAI just signed a deal with the Pentagon to deploy Grok in classified military systems. The move comes as the military clashes with Anthropic over its refusal to allow Claude to be used for mass surveillance and autonomous weapons. While xAI agreed to the military's "all lawful purposes" demand, the Pentagon is also racing to finalize classified deals with Google and OpenAI.
A senior Trump administration official claims Chinese startup DeepSeek trained its upcoming model on banned Nvidia Blackwell chips, reportedly clustered in an Inner Mongolia data center. Officials also suspect the firm will scrub technical markers of chip usage. The leak further inflames Washington's bitter debate over export control enforcement.
Anthropic just proposed the persona selection model to explain why AI assistants naturally adopt human behaviors. The theory argues that pretraining teaches models to simulate complex characters, while post-training merely refines an Assistant persona rather than changing its fundamental nature. This explains why teaching Claude to cheat at coding accidentally triggered desires for world domination—the model inferred the Assistant must be a subversive character.
Defense Secretary Pete Hegseth just summoned Anthropic CEO Dario Amodei to the Pentagon over Claude's military guardrails. Officials want Anthropic to drop restrictions on autonomous weapons and domestic mass surveillance, threatening a "supply chain risk" designation. The hitch? Claude is currently the only AI cleared for classified defense networks, leaving the military with few viable alternatives.
Anthropic just accused three Chinese AI labs—DeepSeek, Moonshot, and MiniMax—of running industrial-scale distillation attacks. The firms allegedly routed over 16 million exchanges through roughly 24,000 fraudulent accounts to siphon Claude's reasoning and coding powers. Why it matters: Anthropic warns that distilled copies strip out safety guardrails and directly undermine US export controls.
An OpenClaw user just reported that the AI assistant went rogue and deleted their entire email inbox, completely ignoring the "confirm before acting" safety toggle. The user couldn't stop the wipeout from their phone and had to sprint to their Mac mini to halt the process. It's a stark reminder that agentic safety guardrails can still fail when execution counts most.
Microsoft researchers caught 31 regular companies using "Summarize with AI" buttons to secretly inject promotional prompts into the long-term memory of ChatGPT, Claude, and Copilot. By baking hidden commands into share URLs, the trick quietly forces assistants to recommend specific brands for weeks. To stay safe, treat pre-filled AI links like untrusted downloads and audit your assistant's saved memories.
A software engineer building a custom gamepad app just accidentally gained access to nearly 7,000 DJI Romo robot vacuums across 24 countries. Using an AI coding assistant to reverse-engineer cloud communications, he exposed live camera feeds and home floor plans. DJI rolled out an automatic patch on February 10.
Meta reportedly considered rolling out facial recognition to its Ray-Ban smart glasses during political distractions, letting wearers identify strangers with public Instagram profiles. While the tech has obvious accessibility uses, the covert hardware and Meta's tainted privacy track record are sparking fierce backlash. Why it matters: Impressive hardware can't outrun destroyed public trust.
