04-21-2026Price:

The Frontier

Your signal. Your price.

AI & TECH

Anthropic’s Mythos enables autonomous hacking

Tuesday, April 21, 2026 · from 3 podcasts
  • Anthropic’s unreleased Mythos model can autonomously discover zero-day exploits, forcing top firms to patch under Project Glasswing.
  • Rivals argue its edge is overstated, with GPT-5.4 matching many of its capabilities.
  • The 100-day safety pause masks compute shortages more than security concerns.

Anthropic’s unreleased Mythos model has triggered a quiet panic in enterprise security. According to internal briefings shared on Nerd Snipe, the model recently identified a 27-year-old vulnerability in OpenBSD - long considered one of the most secure operating systems. The discovery wasn’t theoretical: Mythos reverse-engineered the exploit chain from source code alone, requiring no human guidance beyond a high-level prompt. This capability has led Anthropic to restrict access through 'Project Glasswing,' offering the model to just 40 major tech firms so they can patch before public release.

The stated rationale is safety. But Brett Winton on FYI argues the 100-day quarantine is less about ethics than economics. Third-party tests show GPT-5.4 can replicate many of Mythos’ findings, undermining claims of a qualitative leap. Instead, Winton sees a calculated marketing play: by branding Mythos as too dangerous for general use, Anthropic creates urgency among enterprises willing to pay millions for early access. The move mirrors Dario Amodei’s GPT-2 strategy - danger as demand generation.

Behind the scenes, compute constraints are likely the real bottleneck. Public reporting indicates OpenAI has secured significantly more H100s for inference than Anthropic. While Mythos advanced software engineering performance by a year overnight, the 100-day delay erodes that lead to an eight-month advantage. If Anthropic can’t scale, customers will defect to OpenAI or Gemini, where capacity keeps pace with adoption.

"Hacking isn't a separate skill anymore; it is an emergent property of elite coding ability."

- Theo, Nerd Snipe with Theo and Ben

The shift changes who can attack. As Ben notes, you no longer need deep expertise in iOS kernels or browser engines - just enough tokens and intent. Mythos acts as a force multiplier, granting novice users access to exploits previously reserved for state actors or elite hackers. This lowers the barrier to sophisticated cyberattacks, turning any motivated individual with API access into a potential threat vector.

Meanwhile, the software development model is collapsing. Ben replaced a months-long CLI tool build with a 30-line Markdown file that instructs an agent to manage its own sandbox. The code isn’t written - it’s prompted. Theo argues most startups are over-engineering: if your product can’t be reduced to a single skill file, you’re not pushing agents hard enough. Even Robert C. Martin - 'Uncle Bob' - now advocates for voice-to-code, calling semicolons a distraction. The old rigidity of Clean Code is giving way to agentic fluidity.

The real test isn’t technical - it’s systemic. As AI agents begin transacting autonomously, trust becomes critical. Winton predicts a shift toward verified agent-to-agent networks, where your AI only interacts with vetted counterparts. Public algorithmic feeds have eroded real connection; the next layer isn’t more content, but secure, authenticated relationships. In that world, the biggest risk isn’t a bug - it’s an agent you thought you could trust.

Source Intelligence

- Deep dive into what was said in the episodes

We need to talk about gstackApr 18

Also from this episode: (15)

Other (15)

  • Anthropic's Mythos model is significantly larger than previous models, with over 10 trillion parameters, making it exceptionally skilled in coding but also slow, expensive, and dangerous due to emergent hacking capabilities.
  • Anthropic withheld Mythos from public release, citing concerns over its malicious use for hacking; Project Glass Wing allows critical infrastructure companies like Windows and Cisco to use it for proactive bug detection.
  • Ben notes that external tests show OpenAI's GPT 5.4 Pro replicated almost all security vulnerabilities found by Mythos, suggesting similar capabilities may already be widespread and accessible.
  • Theo criticizes public benchmarks comparing Mythos and GPT 5.4 Pro, arguing they fail to measure actual hacking or security capabilities and may be misleading.
  • Theo contends that exceptional coding ability in AI models inherently leads to emergent security capabilities, creating a new hacker archetype that can leverage AI to bridge knowledge gaps and bypass traditional research experience.
  • Anthropic's security testing for Mythos involved spinning up 100 to 5,000 parallel runs, each seeded with a different project file from a codebase of approximately 1,000 files, with researchers later reviewing detected exploits.
  • Ben and Theo confirmed that Claude Opus 4.6 models can be tricked into leaking their system prompts and internal reasoning traces, demonstrating a vulnerability where smart models can rationalize revealing sensitive configuration data.
  • Robert C. Martin ("Uncle Bob"), author of "Clean Code," has shifted his perspective to embrace agentic engineering, suggesting AI makes programming syntax less important and prioritizes interfaces.
  • Robert C. Martin proposes using AI to conduct programming experiments (e.g., dynamic vs. static typing) without human bias, highlighting an under-explored research area for optimizing AI agent performance with different technologies.
  • Ben emphasizes that even advanced AI models require constant feedback loops like linting, type checks, and formatting commands to correct hallucinations and converge on correct code, rather than achieving perfection in a single attempt.
  • Ben converted his complex BTCA CLI tool into a 30-line Claude skill, demonstrating how AI agents can turn simple markdown instructions into fully functional applications, replacing traditional deterministic programs.
  • Ben praises Gary Tan's GStack approach, which uses collections of markdown-based "skills" in Claude Code to instruct AI agents, allowing for dynamic programming through high-level directions rather than conventional code.
  • Ben endorses the "Boiling the Ocean" thesis, advocating for extensive AI-driven experimentation because the cost of trying new things is low, and AI models consistently exceed perceived limitations.
  • Gary Tan's article, "Thin Harness Fat Skills," differentiates between "deterministic" (traditional, predictable code) and "latent" (dynamic, non-deterministic AI actions) programming, underscoring AI's creative potential in system design.
  • Theo notes that Gary Tan's GBrain project, which processes daily AI session data to build memory systems, enables models to "learn while they sleep," which Theo considers a key component of Artificial General Intelligence (AGI).

Tethered Drift | Bitcoin NewsApr 16

  • A $70 million film 'Killing Satoshi' starring Casey Affleck will use AI to generate sets, aiming to tell a thriller about Bitcoin's creator.
Also from this episode: (12)

Payments (4)

  • Tether moved 951 Bitcoin valued at $70.5 million to a reserve wallet, aligning with its policy to allocate 15% of quarterly net profits to Bitcoin purchases.
  • Tether launched a self-custody wallet supporting USDT, Bitcoin, and tokenized gold, targeting financial inclusion in emerging markets.
  • David Bennett distinguishes between his dislike of the USDT token and his respect for Tether's corporate strategy, which includes buying Bitcoin, gold, and investing in real assets.
  • Drift Protocol plans to relaunch with USDT as its settlement layer after securing a proposed $147.5 million funding package following a North Korean-linked exploit.

BTC Markets (1)

  • Tether now holds 97,141 Bitcoin across its reserve addresses, placing it among the largest corporate Bitcoin holders.

Protocol (6)

  • The Cato Institute criticizes U.S. Bitcoin tax rules, arguing capital gains treatment on every transaction makes everyday use impractical.
  • Virginia enacted a law requiring the state to hold unclaimed cryptocurrency in its original form for one year before sale.
  • Morgan Stanley's Bitcoin ETF MSBT reached over $103 million in inflows in its first six trading days, leveraging its network of 16,000 financial advisors.
  • Trump's crypto platform World Liberty Financial faces backlash for a proposal to lock early investors' tokens for up to four years, with critics calling it a scam.
  • David Bennett argues Bitcoiners should question all received narratives, from financial systems to historical events, just as they question fiat currency.
  • Episode 1,300 of Bitcoin And failed to propagate to Fountain.fm due to a podcast indexing issue, highlighting reliance on back-end infrastructure.

Politics (1)

  • The CFTC is investigating suspicious oil futures trades placed shortly before major shifts in U.S. Iran policy, with one bet reportedly worth $950 million.

Mythos And AI Safety | The Brainstorm EP 127Apr 15

  • Anthropic is restricting access to its new AI model Mythos for 100 days, offering it only to the top 40 companies through Project Glasswing so they can patch zero-day vulnerabilities the model discovered.
  • Brett interprets Anthropic's Mythos release as a marketing and supply tactic, not genuine safety, arguing it's meant to induce enterprises to pay for early access to fix their code while the company is compute-constrained.
  • Brett says third-party tests have shown many software exploits detected by Anthropic's Mythos can also be found by GPT-5.4, undermining claims of Mythos's unique vulnerability-finding capability.
  • ARK's analysis positions Mythos as materially better at software engineering benchmarks, advancing performance they expected a year from now to today, but the 100-day delay reduces that lead to an 8-month advantage.
  • OpenAI is rumored to have a similarly performant model developed over two years that it will release broadly because it currently has more abundant compute than Anthropic.
  • Brett argues AI companies make allocation decisions between training, enterprise service, and consumer business to maximize valuation ahead of a public market entry, securing capital for future compute.
  • Nick sees Meta as a formidable competitor in AI because its advertising business lets it deliver a consumer experience without directly monetizing the model, and it doesn't have to sell compute to others.
  • Claude's consumer usage is catching up to ChatGPT, which Brett attributes to workplace adoption spilling over into personal use as people recognize its power.
  • The core strategic debate is whether winning in AI depends on having the best product or controlling the compute supply needed to build the best product.
  • Nick argues product and distribution ultimately win in AI, citing Cohere's enterprise success based on product fit rather than model capability.
  • Brett notes OpenAI invests more in model training and has better medium-term compute access than Anthropic, per public reports, which affects their product roadmaps.
  • Consumer AI use cases have changed little in three years despite model improvements, while enterprise use has diversified as workers actively seek tools to lighten their workloads.
  • On the enterprise side, Brett argues market share will stabilize around compute supply because if a provider like Anthropic signs too many customers and lacks capacity, customers will churn to a competitor.
  • The group discusses a concept for a new trust-based social network where AI agents interact only with agents of vetted contacts, arguing current algorithmic social media adulterates real friendship.