Alejandro Jimenez Rico @MrKnoot: "𝗔𝗻𝘁𝗵𝗿𝗼𝗽𝗶𝗰 𝗮𝗻𝗻𝗼𝘂𝗻𝗰𝗲𝘀 𝗠𝘆𝘁𝗵𝗼𝘀, 𝘀𝗵𝗮𝗸𝗶𝗻𝗴 𝘁𝗵𝗲..."

United Kingdom

𝗔𝗻𝘁𝗵𝗿𝗼𝗽𝗶𝗰 𝗮𝗻𝗻𝗼𝘂𝗻𝗰𝗲𝘀 𝗠𝘆𝘁𝗵𝗼𝘀, 𝘀𝗵𝗮𝗸𝗶𝗻𝗴 𝘁𝗵𝗲 𝘄𝗵𝗼𝗹𝗲 𝗖𝘆𝗯𝗲𝗿𝘀𝗲𝗰𝘂𝗿𝗶𝘁𝘆 𝗹𝗮𝗻𝗱𝘀𝗰𝗮𝗽𝗲 Anthropic, the company behind Claude, just announced Mythos. Their new frontier model ranks way above Opus 4.6 (and GPT 5.4 Pro, if we're to trust indirect reports) in many intelligence benchmarks. So smart, they say, they've decided NOT to release it, because it seems to be a cybersecurity monster. It finds and exploits software vulnerabilities. It is so good at doing so, that by itself it found zero-day vulnerabilities on massive projects like Linux, FFmpeg and OpenBSD like it's nothing. Vulnerabilities that have been sitting there unnoticed for up to 27 years until Mythos found them! With an almost non-existent harness, Mythos managed to find - completely on its own and with no human steering - THOUSANDS of such vulnerabilities, and was capable of writing a script to take advantage of them. Anthropic has, reportedly, contacted these organisations and advised on the measures to fix them. While deciding NOT to release this model to the general public just yet. Instead, they've decided to spearhead the project 'GLASSWING', in which they gather the top players in the tech industry they can trust to carefully improve infrastructure and software security to get ready for this. Because they might not release this model, but it's clear someone else will also achieve this capability soon enough, and maybe not everyone will do it with good intentions. The report Anthropic has released does smell a little bit like a marketing stunt. "Our model's so good we can't even release it, for humanity's sake. And we're obviously the good guys". In my opinion, there's some of that in it. But I partially buy it. AI has risks, and they're doing, imho, a right job of being careful to mitigate those risks. Even if they're also using their good actions as an excuse to do self-promotion. The truth is, Cybersecurity has changed forever. It has been changing over the last few months, quite frankly. This announcement just made it very public and very obvious. "Security by obscurity" is dead. Breadth and speed is the name of the game now, having an army of agents scrutinising your frameworks to find vulnerabilities. The market for this is immense, and Anthropic is going head-first after it. $CRWD (Crowdstrike Holdings) and $PANW (Palo Alto Networks) are partners in this 40-company coalition that's Glasswing. So I expect the Cybersecurity market to scramble to adapt to this new environment real fast. Every software company now faces accelerated vulnerability discovery. That's deflationary for security costs long-term but creates massive near-term spending pressure. Companies that delay patching infrastructure face existential risk, you can't put a price tag on that. Think of it this way - right now, every company needs to spend more to scan and patch everything before attackers do. But once the dust settles, AI-powered security tools will do in minutes what used to take expensive human teams weeks. The bill goes up now, but it will come down later. Cloud providers are, for sure, allocating a bottomless budget to patch up their systems as fast as they can. I'm talking about the likes of $AMZN (Amazon.com Inc) $GOOG (Alphabet) and $MSFT (Microsoft) here. All of them part of the Glasswing project. And all of them extremely exposed if the Cloud business gets attacked by malicious AI. Buckle up, the tech sector has never been more intensely alive!

Not investment advice. The author may have financial interests in the mentioned instruments.