+ Anthropic Restricts Its Own Model After Finding 27 Year Old Bug
On Monday, Anthropic announced that Claude Mythos, its most capable model to date, will not be released to the public. Instead, it is being made available only to a selected group of more than 40 technology and cybersecurity companies through a program called Project Glasswing.
The reason: Anthropic's internal testing found that Mythos can discover software vulnerabilities that have gone undetected for decades. In one test, the model identified a bug that had been present for 27 years and had never been caught by human reviewers. Anthropic assembled more than 50 cybersecurity defenders to evaluate the model and concluded that unrestricted access would create too much risk.
What it all means: Anthropic is essentially saying the model works so well at finding security flaws that putting it in the wrong hands could cause real damage. The Project Glasswing approach, giving it only to defenders, is a new kind of release strategy. It is also a test of whether the AI industry can self-regulate on capability.
+ OpenAI Tells Congress: Tax the Robots, Fund the Public, Try a Four Day Week
OpenAI published a 13 page policy document titled "Industrial Policy for the Intelligence Age" aimed at shaping how Congress approaches AI legislation. The proposals are sweeping. Among them: a tax on automated labor to replace declining wage-based tax revenue. A public wealth fund that would hold equity stakes in AI companies and distribute returns to citizens. And government-backed experiments with a 32 hour workweek at current pay.
The document also calls for formal worker representation in AI deployment decisions and new safety containment plans for advanced systems. OpenAI plans to host discussions at its upcoming Washington, D.C. workshop opening in May.
What it all means: OpenAI is positioning itself as the company that wants to share the upside, not just capture it. Whether that framing holds up under scrutiny from lawmakers and labor advocates is another story. But the fact that the conversation has moved from "should AI take jobs" to "how do we restructure the economy around AI" tells you where things stand.
+ Meta Launches Muse Spark From Its New Superintelligence Labs
Meta released Muse Spark on Wednesday, the first AI model from Meta Superintelligence Labs. This is the research unit CEO Mark Zuckerberg built after concluding that Meta's Llama models were falling behind OpenAI and Anthropic. To lead it, Meta brought in Alexandr Wang, the former CEO and co-founder of Scale AI, and invested $14.3 billion in Scale for a 49% stake.
Muse Spark accepts voice, text, and image inputs but produces text only. Meta says it is competitive with leading models in areas like understanding images and processing health information, but the company acknowledges it still trails competitors in coding. The model is free to use and will power Meta AI across Facebook, Instagram, WhatsApp, and the Meta AI app.
What it all means: Zuckerberg spent billions on an outside hire and a data labeling deal to catch up. The first result is a model Meta itself says is not best in class yet. Whether this becomes a turning point or an expensive detour depends entirely on what comes next from Wang's team.
+ Amazon CEO Tells Shareholders to "Bet Big" on AI as Spending Hits $200 Billion
Andy Jassy's 2026 shareholder letter laid out the scale of Amazon's AI commitment. The company expects about $200 billion in capital expenditure this year, with the majority going to AWS and AI infrastructure. Jassy said Amazon is "monetizing capacity as fast as we can install it" and identified AI chips, robotics, and low orbit satellites as the primary investment areas.
The letter went further than past communications. Jassy wrote that "every customer experience that we know of today is going to be reinvented" and framed AI not as one bet among many but as the central organizing principle for the company going forward.
What it all means: Amazon is not experimenting with AI. It is rebuilding itself around it, from the warehouses to the cloud to the satellites. When the company that delivers your packages, hosts your streaming, and runs the infrastructure behind half the internet says every experience is about to be reinvented, that is a statement worth taking seriously.
