NewsTechnical

Claude Mythos is Too Dangerous to Use

Department of Product11m 43s

Anthropic has released Claude Mythos Preview, a powerful AI model with exceptional coding abilities that's restricted to select partners due to security risks. The model can find zero-day vulnerabilities and exploits, leading to Project Glasswing - a $100 million initiative to address security concerns before wider release.

Summary

Anthropic has unveiled Claude Mythos Preview, marking the first time their frontier model isn't publicly available due to security concerns. The model is restricted to select partners including Apple, Google, Nvidia, and Microsoft under Project Glasswing, named after glasswing butterflies whose transparency represents both the vulnerabilities discussed and the transparent approach being taken. The model demonstrates impressive coding capabilities, scoring 93.9% on SWE Bench Verify tests and 77.8% on SWE Bench Pro, significantly outperforming predecessors by 24 points. However, these capabilities also enable the model to find software vulnerabilities, including discovering a 27-year-old OpenBSD flaw, a 16-year-old FFmpeg vulnerability, and a Linux privilege escalation bug. The model has reportedly found thousands of zero-day vulnerabilities across major operating systems and web browsers. This development has accelerated the 'token maxing' trend, where companies incentivize employees to use more AI tokens, with Meta creating internal leaderboards ranking employees by token consumption. Google's product director suggests that go-to-market skills are becoming crucial as AI democratizes building capabilities. Industry data shows GitHub commits have increased 14x year-over-year to 275 million weekly, while company AI spending has grown 4x, with median companies dedicating 15% of software budgets to AI tools. Despite concerns about job displacement, tech job openings are up 30% this year, though 52,000 tech job cuts were announced with AI cited as a leading reason.

About this episode

Anthropic has revealed Claude Mythos Preview — a new frontier model it's calling too powerful for public release. Instead, it's being made available exclusively to a select group of partners including Apple, Google, Microsoft, and NVIDIA under an initiative called Project Glasswing. We also cover Meta's internal "Claudeonomics" leaderboard turning token usage into office status, new data on GitHub commits exploding 14x year-on-year, Perplexity's ARR surging past $450M, and Google's Product Director making the case that Go-to-Market is becoming the essential skill in the AI age. ➡️ Subscribe for weekly product briefings and more analysis: https://departmentofproduct.substack.com Follow on Substack Notes: https://substack.com/@richholmes 🔗LINKS Project Glasswing announcement — https://www.anthropic.com/glasswing Claude Mythos Preview system card — https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf Felix Rieseberg on Mythos being a "step function change" — https://x.com/felixrieseberg/status/2041586309966524919 Simon Willison on why the pause "sounds necessary" — https://simonwillison.net/2026/Apr/7/project-glasswing/ Ethan Mollick on security risks — https://x.com/emollick/status/2041578945531830695 Meta's internal AI token leaderboard — https://www.theinformation.com/articles/meta-employees-vie-ai-token-legend-status?rc=77sebk Jensen Huang on token spending — https://embed.businessinsider.com/jensen-huang-500k-engineers-250k-ai-tokens-nvidia-compute-2026-3 Zapier's AI fluency framework — https://x.com/wadefoster/status/2038979630590509553 Linear's COO on token-maxxing — https://x.com/cjc/status/2041299419845599489 Google's Product Director on GTM as the essential skill — https://x.com/jacalulu/status/2041160452672004189 The SaaS chat bar trend — https://x.com/rabi_guha/status/2040082295563169852 Simon Willison on GitHub commits — https://simonwillison.net/2026/Apr/4/kyle-daigle/ Ramp: monthly AI spend grew 4x — https://ramp.com/3-steps-to-manage-ai-spend Perplexity ARR tops $450M — https://ca.finance.yahoo.com/news/perplexity-arr-tops-450m-pricing-132500539.html AI and software engineering jobs — https://www.businessinsider.com/ai-isnt-killing-software-coding-jobs-booming-trueup-2026-4 Substack article on new product development processes - https://departmentofproduct.substack.com/p/the-new-product-development-operating

Key Insights

  • Anthropic argues that Claude Mythos Preview's coding abilities are so advanced that it can surpass all but the most skilled humans at finding and exploiting software vulnerabilities
  • The speaker claims that Project Glasswing represents a proactive response to the reality that such powerful capabilities will eventually spread beyond actors committed to deploying them safely
  • Meta has built internal leaderboards that rank 85,000 employees by AI token consumption, turning token usage into a measure of engineering prestige with badges like 'Token Legend'
  • Google's product director contends that go-to-market has become the essential skill as AI democratizes building capabilities, shifting the question from 'can you build' to 'should you build'
  • The analysis shows that GitHub commits have exploded 14x year-over-year while median companies now dedicate nearly 15% of their total software budget to AI tools

Topics

AI SecurityCode Vulnerability DetectionToken MaxingAI Development TrendsTech Industry Employment

Transcript

Anthropic says its latest model beats all other AI coding benchmarks, but for now it's limiting its release to a handful of partners because of security concerns. We'll take a look at what this new Claude Mythos model is capable of, why Anthropic is worried, and what product development teams need to know right now. Plus, why Meta has built a leaderboard to incentivize so-called token maxing. New data shows how perplexity computer may have transformed the fortunes of the company, and Google's product director reveals what she thinks is the essential skill in the AI age. So stay tuned for all of that and more, and as always, if you do enjoy the briefing, then hit the like and…

Full transcript available for MurmurCast members

Sign Up to Access

More from Department of Product

Get AI summaries like this delivered to your inbox daily

Get AI summaries delivered to your inbox

MurmurCast summarizes your YouTube channels, podcasts, and newsletters into one daily email digest.