Anthropic is bringing its controversial Mythos cybersecurity model to all customers "in the coming weeks," the company confirmed Thursday, even as it launched a modest upgrade to its Opus line that prioritizes honesty over raw capability.
Mythos, Anthropic's large language model with advanced cybersecurity capabilities, has been restricted to a small number of organizations since its unveiling in early April. Reuters reports that as part of Project Glasswing, Amazon, Microsoft, and Apple are permitted to use Mythos for cybersecurity purposes. The model's potential impact has raised concerns among executives and world leaders.
Alongside that news, the company released Claude Opus 4.8 today, its second Opus upgrade in six weeks. The previous version, Opus 4.7, arrived on April 16.
Opus 4.8 is available globally at the same price as its predecessor.
Anthropic describes Opus 4.8 as having "sharper judgement, more honesty about its progress, and the ability to work independently for longer than its predecessors." Early testers report the model is roughly four times less likely to let coding flaws pass unremarked, and is more inclined to flag uncertainties rather than make unsupported claims.
"A general problem with AI models is that they sometimes jump to conclusions, confidently claiming to have made progress in their work despite the evidence being thin," Anthropic said.
Benchmarks show Opus 4.8 scoring 69.2% on SWE-Bench Pro, outperforming GPT-5.5 and Gemini 3.1 Pro on that test. Agentic coding scores rose from 64.3% to 69.2%, multidisciplinary reasoning with tools jumped from 54.7% to 57.9%, and knowledge work scores increased from 1753 to 1890.
GPT-5.5 still leads on terminal-coding benchmarks.
Opus 4.8's fast mode now runs at 2.5x the speed and costs three times less than prior models.
Anthropic also introduced three new features alongside the model. Dynamic workflows, available in research preview, lets Claude plan work and run hundreds of parallel subagents in a single session, capable of completing codebase-scale migrations across hundreds of thousands of lines of code. A new effort control in Claude.ai and Cowork lets users choose how much compute Claude devotes to a response, lower settings return faster answers and conserve rate limits, while "extra" and "max" settings spend more tokens on difficult tasks.
The Messages API now accepts system entries inside the messages array, letting developers update Claude's instructions mid-task without breaking the prompt cache. The dual announcement, a safety-focused Opus upgrade alongside the impending Mythos broad release, highlights Anthropic's strategy of maintaining public trust through demonstrated honesty improvements while pushing the boundaries of what its most powerful models can do.
