NEW Claude Mythos and ChatGPT Updates!
Anthropic accidentally leaked details about their most powerful model ever, Claude Mythos, while OpenAI shut down Sora to focus resources on a new model codenamed Spud. Both developments represent significant shifts in AI capabilities, with Mythos raising cybersecurity concerns and Spud promising novel capabilities that could 'accelerate the economy.'
Summary
In March 2026, two major AI developments occurred that signal significant changes in the industry. First, Anthropic suffered a security breach that exposed nearly 3,000 internal files, including draft documentation for Claude Mythos, their most powerful unreleased model. The leaked documents revealed that Mythos represents a new fourth tier above their current Opus model, with dramatically improved performance in software coding, academic reasoning, and cybersecurity capabilities. However, the cybersecurity capabilities are causing concern, as Anthropic's own documents describe the model as posing 'unprecedented cybersecurity risks' and being 'far ahead of any other AI model in cyber capabilities.' Chinese state-sponsored groups have already misused previous Claude models to infiltrate approximately 30 organizations. Due to these safety concerns and high computational costs, Mythos is currently only available to a small group of early access customers, primarily cybersecurity defense organizations, with no confirmed public release date though Q3 2026 seems realistic. Simultaneously, OpenAI completed pre-training on a new model codenamed Spud around March 24th, 2026. CEO Sam Altman described it as potentially able to 'accelerate the economy' and featuring completely novel capabilities not seen in existing models. To support Spud's development, OpenAI made the dramatic decision to shut down Sora, their video generation model, abandoning licensing deals with Disney and redirecting those computational resources. The company has undergone significant restructuring, with Altman stepping back from safety oversight to focus on fundraising and data centers, while the product division was renamed 'AGI deployment.' OpenAI is developing a unified desktop super app combining ChatGPT, their coding agent, and a browser called Atlas, all powered by Spud. Both developments indicate the AI industry is entering a new phase of capability and competition, with release timelines expected within weeks for Spud and potentially Q3 2026 for Mythos.
Key Insights
- Anthropic's security breach exposed Claude Mythos documentation, which the company describes as a 'step change in AI performance' and their 'most capable model ever built'
- Claude Mythos poses 'unprecedented cybersecurity risks' according to Anthropic's own leaked documents, being 'currently far ahead of any other AI model in cyber capabilities'
- Chinese state-sponsored groups previously used Claude to infiltrate roughly 30 organizations including tech companies, financial institutions, and government agencies
- OpenAI shut down Sora and abandoned a verbal licensing agreement with Disney to redirect computational resources toward their new model Spud
- Sam Altman told employees that Spud features 'a completely novel capability not seen in any existing model' and could 'really accelerate the economy'
Topics
Full transcript available for MurmurCast members
Sign Up to Access