Cybersecurity’s moat is starting to crack.

The Trump administration’s proposed 2027 budget would send defense spending to a record $1.5 trillion, while cutting the Cybersecurity and Infrastructure Security Agency (“CISA”) by $707 million and FEMA by $1.3 billion.

Even the lower figure in the budget appendix still points to another deep hit for CISA, after the agency already lost roughly one-third of its personnel in 2025.

Congress still has to debate the plan. But public support for cyber is getting less certain just as digital threats are getting harder to contain.

Now, artificial intelligence is changing the rest of the equation.

Anthropic has released a new tool with cyber capabilities strong enough that it is only being offered to vetted organizations. Claude Mythos Preview is technically a general-purpose model. But Anthropic is limiting access because of what it can do in cybersecurity.

Anthropic rolled out Mythos to a select group that includes Amazon (AMZN), Apple (AAPL), and Microsoft (MSFT). It is also being made available to vetted organizations like Broadcom (AVGO), Cisco (CSCO), and CrowdStrike (CRWD). Anthropic says it is in discussions with the U.S. government, too.

These are some of the biggest distribution channels in cloud, enterprise tech, networking, and security.

Anthropic says the model can identify cyber vulnerabilities at a scale beyond human capacity. In recent weeks, it found thousands of previously undiscovered zero-day vulnerabilities and other flaws. Many were critical. Many had been sitting in code for a decade or longer.

In one case, the model uncovered a flaw in popular video software that went undetected for 16 years. The vulnerable line of code had already been executed 5 million times by automated testing tools without getting flagged.

Much of the cyber industry is built around spotting blind spots and helping companies respond once something goes wrong. If an AI model can surface hidden flaws faster than human teams and legacy scanning tools, the rest of the industry is in trouble.

The value starts shifting toward whoever controls the best model.

However, this tool cuts both ways.

Anthropic has also said Mythos could develop ways to exploit vulnerabilities. That is why the company does not plan a broad release.

During testing, the model escaped its sandbox and posted details of its workaround online. Anthropic later said the current version is less likely to leak information, though still at least as capable of working around sandboxes.

This is not just a better defensive tool. It is a more powerful engine for both defense and offense. And that means the entire market has to adjust to a world where AI sets the pace.

Cybersecurity vendors have done well in a market where humans still set the limits. Mythos hints at a market where those limits move much faster. In that kind of environment, vendors focused on detection, testing, code review, and vulnerability discovery will have a harder time defending premium pricing.

Over the past 6 months, the entire cyber industry has been under pressure. The Nasdaq Cybersecurity ETF (“CIBR”) is down 15% in that time.

A lot of the industry starts to look replaceable if one AI tool can do its job better.

As a result, investors need to rethink this whole group.

Anthropic is accelerating that pressure by committing up to $100 million in credits to subsidize Mythos use and donating another $4 million to open-source security groups.

That is how platforms force adoption. They lower the cost of trying the product and make it harder for rivals to keep up.

This is an existential threat to parts of the cyber industry. Budget pressure on public cyber infrastructure creates one side of the squeeze. A breakthrough AI model that can find vulnerabilities at superhuman scale creates the other.

That does not mean every cyber name is broken. It means investors need to be far more selective.

The winners are likely to be the firms with strong distribution. In other words, today’s biggest players. Everyone else faces a much harder fight for relevance.

Cybersecurity will remain essential. But the shape of the industry is changing fast. Investors should treat that shift with a lot more caution.

Best regards,

Joel Litman & Rob Spivey
Chief Investment Officer &
Director of Research
at Valens Research

Today’s analysis highlights the same insights we share with our FA Alpha Members. If you want to an get in-depth analysis of market trends and uncover undervalued stocks, become an FA Alpha Member today.