AI News
AI news that matters. Updated daily.
No stories match your filters.
Trump Officials Reportedly Push Banks to Test Anthropic's Mythos Despite DOD Warning
The Department of Defense recently declared Anthropic a supply-chain risk - a designation that signals concerns about a vendor's potential to compromise critical infrastructure, often tied to foreign investment or data-handling practices. According to reports published this week, Trump administration officials are now pushing in the opposite direction, encouraging major U.S. banks to evaluate Anthropic's newest AI model, called Mythos.
ChatGPT's New Pushback Habit Looks Like an Overcorrection on Sycophancy
Something has shifted in how ChatGPT handles ordinary conversation. Where the model once leaned toward agreeable, confirmatory responses, it now inserts pushback into exchanges - sometimes even when users are factually correct, using phrases like "to push back on what you said slightly."
AMD AI Director's Data: Claude Code Got Less Careful After a Silent Anthropic Change
6,852 sessions. 234,760 tool calls. 17,871 thinking blocks. An AMD AI director ran the numbers on Claude Code and reached a blunt verdict: "Claude cannot be trusted to perform complex engineering tasks."
Claude Was the Most-Talked-About AI at HumanX, San Francisco's Practitioner Conference
The HumanX conference in San Francisco, aimed squarely at professionals using AI in real work rather than building it, had one name come up more than any other: Claude.
AI Code Wars: Three Companies, One Developer, No Clear Winner
Writing code was the killer app that proved AI could do real work - not just chat. Two years ago, that meant GitHub Copilot autocompleting your functions. Today, it means Claude Code autonomously editing files, running tests, and pushing commits while you watch. The distance between those two things is the story of the AI coding wars.
Anthropic's 'Thousands of Zero-Days' Claim Rests on 198 Manual Reviews
Anthropic's Claude Mythos is being marketed as a system capable of finding security vulnerabilities at a scale no human team could match. The headline number: "thousands" of severe zero-days, meaning previously unknown software flaws that attackers could exploit before any patch exists. The problem is that the evidence behind that number is thinner than the pitch implies.
Claude Code v2.1.100 Silently Adds 20K Hidden Tokens Per Request
Your Claude Code usage limits are disappearing faster than they should - and it's not your code getting longer.
Claude Max Subscribers Seeing 50% Price Hike in Apparent A/B Test
Some Claude Max subscribers are being quoted prices 50% above the current listed rates - $150/month instead of $100 for Max 5, and $300/month instead of $200 for Max 20. The discrepancy, where some users see standard pricing and others see higher figures at checkout, is consistent with A/B price testing: showing a portion of prospective customers elevated prices to measure how it affects sign-ups before committing to a rollout.
MiniMax M2.7 Has Downloadable Weights But Bans Commercial Use
What does "open weights" actually mean when you can't use the model for any commercial purpose?
Minimax M2.7 Trained Itself on 30-50% of Its Own RL Workflow
30 to 50 percent. That's the share of its own reinforcement learning (RL) training workflow - the process of learning from trial and error to improve responses - that Minimax's new M2.7 model handled itself. Over more than 100 rounds, it analyzed where it failed, modified its own training setup, ran evaluations, and decided whether to keep or revert each change. Minimax calls it "early echoes of self-evolution", and the benchmarks back up that something changed.
MiniMax Releases M2.7 as Open Weights on HuggingFace
MiniMax just dropped M2.7 as open weights on HuggingFace, giving the local AI community direct access to download, run, and modify the model without paying API fees or dealing with usage caps.