Anthropic's 'Mythos' Model Leak Reveals a Step Change in AI Capabilities
The biggest story shaking the AI world this week centers on Anthropic, which inadvertently exposed internal documents about its most powerful model yet — codenamed "Capybara" and publicly dubbed Claude Mythos. A misconfigured content management system left nearly 3,000 internal files publicly accessible, including draft blog posts describing the model's capabilities.
According to the leaked documents, Mythos represents an entirely new tier above Anthropic's existing Opus models. An Anthropic spokesperson confirmed the model represents "a step change" in performance, scoring "dramatically higher" than Claude Opus 4.6 on software coding, academic reasoning, and — most notably — cybersecurity benchmarks.
What's drawing the most attention is the model's reported ability to autonomously identify and patch vulnerabilities in its own code. The draft describes Mythos as "currently far ahead of any other AI model in cyber capabilities" and warns it "presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders."
The implications were immediate. Cybersecurity stocks including CrowdStrike, Palo Alto Networks, and Fortinet sold off as investors assessed what more capable AI cyber tools could mean. Bitcoin also dropped to $66,000 amid broader risk-off sentiment.
Anthropic is reportedly restricting early Mythos access to cyber defense organizations and has privately briefed top government officials on the risks. No timeline for general availability has been announced, with the company noting the model is "very expensive to serve" and requires significant efficiency improvements before a wider rollout.
OpenAI Shuts Down Sora Video Generator
In a surprising move, OpenAI has discontinued Sora, its AI video generation product. The company is redirecting the computational resources previously allocated to Sora toward higher-priority projects — a clear signal that GPU allocation remains a zero-sum game even for the world's best-funded AI lab.
The decision comes as OpenAI focuses on its core GPT-5.4 model family, which recently launched with a 1-million-token context window and autonomous multi-step workflow capabilities. On the OSWorld-V benchmark, GPT-5.4 scored 75%, slightly above the human baseline of 72.4% — a milestone that positions the model as a genuine autonomous digital coworker rather than a simple chat tool.
Sora's shutdown also coincides with intensifying copyright litigation from major record labels against AI-generated content, with lawsuits against competitors like Suno alleging unauthorized use of training data from YouTube. The legal landscape for generative media may have factored into OpenAI's calculus.
Source: TechStartups
Mistral Secures €830M to Build European AI Infrastructure
France's Mistral has closed an €830 million debt financing round — its first major debt raise — to purchase 13,800 Nvidia chips and construct a major data center near Paris. The seven-bank financing deal represents one of the largest European AI infrastructure investments to date.
The move is part of Mistral's broader ambition to build European AI sovereignty through owned capacity and local infrastructure, with a target of 200 megawatts of compute across Europe by the end of 2027.
This comes on the heels of Mistral's successful open-source release of Mistral Small 4, a 22-billion parameter model under the Apache 2.0 license that has outperformed several closed models three to five times its size on reasoning benchmarks. The model runs on a single A100 GPU, making it viable for on-premise enterprise deployments — exactly the kind of use case Mistral's new infrastructure will support.
Source: TechStartups
Google DeepMind Launches Lyria 3 AI Music Generator
Google DeepMind has unveiled Lyria 3, its latest AI music generation model, now available in beta within the Gemini app. The model generates 30-second music tracks with auto-generated lyrics from simple text prompts, supporting multiple languages.
Lyria 3 enters an increasingly competitive AI music space, but also an increasingly litigious one. The launch comes as major record labels escalate copyright lawsuits against AI music generators, setting up a collision between creative AI capabilities and intellectual property rights that will likely define the sector for years to come.
Source: OpenTools AI News
Eli Lilly Bets $2.75 Billion on AI Drug Discovery with Insilico Medicine
Pharmaceutical giant Eli Lilly has signed an exclusive licensing agreement with Insilico Medicine worth up to $2.75 billion, including $115 million upfront. The deal marks one of the largest commitments to AI-driven drug discovery to date, signaling that the pharmaceutical industry is moving from pilot programs to serious capital deployment in AI-assisted development.
Insilico Medicine uses AI to accelerate every stage of the drug discovery pipeline, from target identification to clinical trial design. The partnership validates the thesis that AI can meaningfully compress drug development timelines — a promise the industry has been betting on for years that is now attracting billion-dollar commitments.
Source: TechStartups