Yes, the one now notorious for its seemingly uncanny capabilities in figuring out cybersecurity attack vulnerabilities. Apparently it’s now showing capabilities that are deemed a ‘must have’ for countless companies as protection from hacking threats and liabilities. A new area of ‘product-market-fit’ for Anthropic beyond AI Coding.
It’s resulting in AI capabilities and pricimg far removed than the regular, mostly free AI regular mainstream customers are getting via Google Gemini, OpenAI ChatGPT and the like.
Even the folks paying $20 to $200/month for AI chatbot and agentic services are increasingly not getting the ‘REAL AI” that top developers, corporate customers and now cybersecurity executives are paying. It can easily cross a million dollars a month and far more.
The head of a top VC incubator discussed his personal AI bill already crossing seven figures a month.
“Anthropic’s Mythos identifies critical vulnerabilities five times faster, Palo Alto Networks said.”
“Mythos testing quickly consumes millions of dollars of tokens for early adopters.”
“Companies say they are planning budget increases for advanced AI cybersecurity tools.”
“When Palo Alto Networks earlier this year began testing Anthropic’s Claude Mythos to comb through its own source code, it didn’t take long to see the future of cybersecurity. It also didn’t take long to see what that future would cost.”
“The model found more than two dozen critical vulnerabilities in around three weeks, roughly five times what the company would typically find using existing tools, said Sam Rubin, senior vice president of Palo Alto Networks’ threat intelligence arm. The company also “very quickly” burned through more than $1 million worth of tokens using Mythos, he said.”
“Another Mythos tester said using the model even for just a couple weeks can cost millions of dollars.”
“Anthropic has been subsidizing early Mythos testers, meaning Palo Alto Networks didn’t need to pick up the tab for the tests. But corporate cybersecurity executives say they are preparing to boost their spending budgets for Mythos and other cutting-edge AI that will reshape how companies defend themselves against hacks.”
“Anthropic itself says Mythos, which it hasn’t broadly released to the public over fears the AI would be used to facilitate attacks, will be about six times more expensive per token compared to the most advanced Anthropic model that’s widely available, Opus. And Opus is already on the pricier end of the market, as many companies recently found out. (Tokens are the words or parts of words an AI model processes to produce an answer.)”
“However, Mythos performs much better on sophisticated cybersecurity tasks compared to Opus, according to a UK agency that has tested the models. That suggests customers are likely only paying about twice as much for Mythos.”
And companies are starting to rationalize the extraordinary AI expense for safer cybersecurity.
“Other corporate security leaders also said paying up for Mythos is necessary, given that major breaches can cost companies tens or hundreds of millions of dollars in legal fines and other payouts to affected customers and partners.”
“These executives say they expect to spend heavily on such models in the year ahead to analyze their companies’ code and find vulnerabilities before hackers do. They’re also evaluating whether to buy additional cybersecurity tools to defend against specialized cyberattacks in which malicious hackers use AI to break into companies, a trend that’s on the rise. (Anthropic says it’s developing features to prevent hackers from using Mythos.)”
For now, Anthropic’s Mythos has a mythic stature developing around its capabilities and their being ‘worth it’.
“he hype around Mythos is “causing all organizations to reevaluate their security posture,” said Ryan Downing, chief information officer of enterprise business solutions at investment firm Principal Financial Group, which has nearly $800 billion in assets under management. “Before, there was always time between a vulnerability being discovered and exploited, and a lot of processes were built on that assumption, but that’s not true anymore.”
And that of course for Anthropic, is a good thing tactically, around its upcoming mega-AI IPO just filed.
“The upcoming offering could further propel Anthropic, which has rapidly leapfrogged archrival OpenAI in revenue on the strength of its AI for coding and other workplace tasks, including automating some low-level cybersecurity work. While Anthropic plans to charge a hefty premium to Mythos customers over existing models, the company has said the model is extremely costly to operate. It isn’t clear how that will impact Anthropic’s gross profit margins, which lately have been strong.”
OpenAI is playing catchup with its Codex and GPT 5.5.
“OpenAI has similarly shared its latest cybersecurity-focused model, GPT-5.5-cyber, with a handful of companies for testing rather than releasing it widely. Mythos is somewhat more advanced than OpenAI’s model, the U.K.’s AI testing agency has said.”
The markets are absorbing the logic for these higher AI costs.
“Paying up for new AI cybersecurity protection could make financial sense, some executives say. The FBI recorded nearly $21 billion in losses in the U.S. last year stemming from cyberattacks, up from $16.6 billion the year prior. Ransoms paid by companies that fell victim to hackers in recent years have frequently reached the tens of millions of dollars in individual incidents, though many payments aren’t made public.”
“Even without Mythos, companies are spending more on AI for security.”
And they’re finding ways to possibly stretch the AI budgets.
“Early adopters are already developing systems that could minimize their Mythos use. UiPath initially developed a set of instructions for the models, known as a skill, that used around 150,000 tokens every time the AI was “warming up” before actually conducting the investigation, he said. That would cost several dollars per task with leading models from Anthropic or OpenAI, which could quickly add up.”
“Similarly, Palo Alto Networks reduced its Mythos usage by offloading some of the work to cheaper models, Rubin said. The firm has set up a system where Mythos creates the plan for breaking into software and then directs a cheaper model like Opus 4.7 to execute it, driving down costs.”
“Mythos has escalated this in importance to where the board is asking, CFO is asking, CEO is asking how they need to make sure their security program is prepared for this,” Rubin said. “And the CFO is much more likely to listen to the CISO in terms of preparedness, so I do think we’ll see budget increases.”
The whole piece is worth a fuller read to understand the current ‘velvet rope J-Curve’ pricing dynamics in enterprise AI. First with AI Coding and now with cybersecurity AI budgets.
And the broader current phenomenon of at least two classes of AI capabilities for those who don’t pay, pay some, and pay a whole lot already. It’s unclear whether these K-curves will get more exaggerated or not as AI Data Center compute capacity gets built for hundreds of billions per year.
It’ll be interesting to see what other enterprise AI applications could next justfiy higher AI budgets this AI Tech Wave. Stay tuned.
(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here)
Keywords CES 2023, AI technology, gadgets, smart home, robotics, innovation, trends, technology review, consumer electronics Summary In this episode of...
