Inside Anthropic's May 2026 Pivot: Claude Opus 4.8, the $965B Valuation, and Unleashing Mythos

Anthropic orchestrated a major strategic shift on May 28, 2026, bundling the launch of its latest flagship model with a road map for its most powerful—and controversial—AI system. The company introduced Claude Opus 4.8, a meaningful upgrade to its public-facing AI assistant, but the bigger news was the decision to begin rolling out the Mythos-class models broadly. This move reverses a months-long policy of keeping the exceptionally capable cybersecurity model under strict lock and key within Project Glasswing. The announcements were overshadowed only by news of a staggering $65 billion funding round that redefines the company's financial firepower .

Claude Opus 4.8: A More Honest and Autonomous Collaborator

Claude Opus 4.8 replaces Opus 4.7 immediately, retaining the same pricing of $5 per million input tokens and $25 per million output tokens . Anthropic positions it as a "more effective collaborator," fine-tuned for complex reasoning, long-horizon agentic coding, and high-autonomy tasks across a 1-million-token context window .

Several improvements stand out. On the capability side, Opus 4.8 delivers better results on agentic coding, multidisciplinary knowledge work, and agentic computer use . A new Dynamic Workflows feature in Claude Code allows the model to orchestrate hundreds of parallel sub-agents for codebase-scale migrations spanning hundreds of thousands of lines . Users also get configurable effort levels—high, xhigh, and max—and a fast mode that operates at 2.5× speed while being three times cheaper than previous fast modes .

Reliability received a major upgrade. Opus 4.8 has been trained to flag uncertainty in its own work more effectively. It is roughly four times less likely than its predecessor to let a flaw in its own code pass without flagging it, addressing a common frustration where advanced models confidently ship broken output . Anthropic describes this as a breakthrough in model honesty for software engineering .

Crucially, while Opus 4.8 is a stronger autonomous agent, it is built on the same ASL-3 (AI Safety Level 3) deployment and security standards as the rest of the Claude Opus 4 family. These protocols include stringent safeguards that automatically block requests indicating prohibited or high-risk cybersecurity uses, as well as robust security controls to protect model weights from theft by sophisticated non-state actors .

The Mythos Turning Point: From Secrecy to a "New Class of Model"

The story of Claude Mythos Preview is what truly separates this launch day from routine model updates. In April 2026, Anthropic revealed the model's existence but explicitly refused to release it publicly. The core issue: Mythos Preview demonstrated autonomous discovery and exploitation of zero-day vulnerabilities across every major operating system and web browser . The UK AI Security Institute (AISI) confirmed that Mythos Preview solved 73% of expert-level capture-the-flag (CTF) cybersecurity challenges—a staggering result that left previous models in the dust .

To mitigate the obvious risks, Anthropic created Project Glasswing, a controlled initiative with about 50 partner organizations—including AWS, Apple, Google, Microsoft, CrowdStrike, and JPMorgan Chase—who used Mythos Preview exclusively for defensive vulnerability discovery . The results were historic. Partners found more than 10,000 high- or critical-severity vulnerabilities across the most systemically important software in the world, with most organizations discovering hundreds of such flaws each within their first month of access . Of those findings, 6,202 were classified as novel, previously unknown zero-day vulnerabilities . The discoveries spanned critical infrastructure and included deeply embedded flaws, such as a 27-year-old vulnerability in OpenBSD, one of the most security-hardened operating systems available .

On May 28, Anthropic declared the experiment a success and announced it was preparing to release the Mythos-class models as a "new class of model" to all customers in the coming weeks . The Glasswing results provided evidence that with proper guardrails, these models could be handled safely for defensive cyber work. Industry analysts report that Claude Mythos 1 is expected in June 2026, launching alongside other major models like Gemini 3.5 Pro and Grok 5 .

The $965 Billion Context Behind the Launch

The product releases come with a financial jolt. On the same day, Anthropic announced it had closed $65 billion in Series H funding at a post-money valuation of $965 billion, more than doubling its previous $380 billion valuation from February . The funding round was led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital, with significant commitments from Blackstone, D.E. Shaw, DST Global, and Fidelity .

The capital officially makes Anthropic the world's most valuable AI startup, surpassing OpenAI's last confirmed valuation of $852 billion . Anthropic stated the funds will be channeled into expanding computing capacity to meet growing enterprise demand for Claude and to scale products globally .

A Collision Course for Cybersecurity and Ethical Hacking

The decision to push Mythos into broad release is not without loud detractors. At the heart of the debate is the "dual-use" dilemma: the same reasoning power that makes Mythos exceptional at finding vulnerabilities to patch also makes it the ultimate offensive toolkit . Analysts warn that broad access could arm malicious actors with a tool that can autonomously chain multiple exploits to bypass operating system sandboxes and browser renderer protections .

The traditional bug bounty and ethical hacking model now faces an existential question. If a single model can autonomously uncover thousands of zero-days, the role of the human security researcher shifts from discovery to triage and containment . Anthropic's containment strategy—using Project Glasswing to filter access to 50 vetted enterprises—has drawn both praise and criticism. Supporters argue it prevented catastrophic leakage while proving the model's defensive value; critics argue that keeping such a tool away from the wider security community slows down the global patching process .

The May 28 announcements mark the end of an experiment and the beginning of a new phase. Anthropic is betting that the defensive benefits of a widely available Mythos-class model outweigh the offense risks, a wager that will be tested as soon as it ships.