🤔 All the talk about the "AI bubble", "who's going to review the slop from agents" and "the meatbag is cheaper" - these are temporary debates by people who miss the main point. Corporations won't be talked into AI coding by CEO motivational speeches - good old Benjamin Franklin will make them.
The signals are already coming - I read the Google Threat Intelligence report for May 2026:
They're tracking agentic LLM systems that just roam the internet and hack. In May 2026 they documented the first real case - AI created a zero-day exploit in a live attack. Microsoft counts 600M+ cyberattacks a day on its customers, ransomware grew 2.75x year over year - and that's before Mythos-class models became widely available
Anthropic built Mythos - plug it into Claude Code and you get the best coding agent and the best IT security expert on the planet at the same time. It found thousands of zero-days in every major OS, every browser. OpenAI's counterpart is the Cyber series with limited access through Trusted Access. You can't get into either one off the street - you need a company audit, and most developers will never pass that filter
Who's already in the Project Glasswing club:
📌 Amazon, Apple, Microsoft - key partners of the program
📌 Cisco, CrowdStrike, Palo Alto Networks - working with Mythos on defense
📌 $100M in credits Anthropic put into the program
📌 Linux Foundation - protecting critical open-source infrastructure
How this breaks the market when Mythos goes open source:
⚡️ Month one - attackers launch agent swarms, automated hacking of everyone who isn't an Anthropic or OpenAI partner
⚡️ The first big corporation goes down - ransomware across the whole corporate network, no longer hypothetical
⚡️ After that event - "the AI agents debate" in IT departments turns into a question of business survival
⚡️ Those in the club - a defender agent plus the best coding tool in one package
⚡️ Those outside - 24/7 attacks and a queue to Anthropic/OpenAI that might not even let them in
Chinese labs lose even more here. Their main way to get Mythos-level capabilities is distillation through American models. Anthropic has already named them publicly: DeepSeek, Moonshot and MiniMax pushed 16M+ requests through 24,000 fake accounts to steal the logic of agentic reasoning and coding. "Access by invitation list" to Mythos and Cyber seriously slows that route down - an open-source counterpart will show up, but it'll take more than a year
💼 Anthropic and OpenAI haven't even started making the real money yet - the peak will come after the first major corporate catastrophe from an AI attack
⭐️ Mikko Hyppönen, Chief Research Officer, WithSecure:
"AI doesn't make criminals smarter - it makes them faster. In cybersecurity speed is everything: defenders need days to patch, attackers need minutes to exploit"
💭 I'm waiting for that moment - not because I want anyone to get hurt, but because only after the first real corporate catastrophe from an AI agent attack will the arguments about "do we even need AI coding" end by themselves. A hand-written backend without an AI defender is a target, not a topic for an IT discussion