Silicon Valley Is Done Playing Nice With Anthropic — And a $1 Billion Cybersecurity Deal Just Changed Everything

Anthropic just had what might be the best month any AI company has ever had — new flagship models, surging enterprise contracts, and its cofounders signing a landmark AI safety letter alongside the chief scientists at OpenAI, Meta, and Google. And Silicon Valley is furious about it. Meanwhile, a $1 billion acquisition dropped this week that I think will matter far longer than the Anthropic drama. Let me break both down.

Why Silicon Valley Is Turning on Anthropic

The Wall Street Journal reported this week that Silicon Valley stakeholders are increasingly frustrated with Anthropic, and the criticism is coming from three distinct directions. First, competitive tactics: Anthropic has launched four flagship models in 2026 and pushed hard into enterprise sales, which critics see as cutthroat behavior wrapped in a safety-first brand. Second, guardrails: Anthropic's models refuse more things than competitors, and for developers doing legitimate work that requires flexible AI behavior, that's a genuine friction point. Third, and most pointed: Anthropic's refusal to back open-weight AI models.

That last one landed with real force this week. Nvidia CEO Jensen Huang's open-weights letter — urging Washington against restricting open AI models — doubled to 50 signatories in a single day, including OpenAI and Google. Anthropic declined. Amazon also didn't sign, but Anthropic's position drew fire specifically because it has spent considerable energy advocating for AI oversight while keeping its own models tightly closed. To the open-source community, that looks like using safety as a competitive moat.

The Counterpoint: Anthropic Is Winning

Here's the tension that makes this story genuinely interesting: the same week Silicon Valley is griping about Anthropic, Anthropic's numbers and reputation are as strong as they've ever been. Enterprise adoption is up sharply. The safety-forward positioning keeps winning large institutional customers who won't trust less careful alternatives. And Anthropic's cofounders, Jack Clark and Jared Kaplan, just signed the landmark AI pacing letter alongside chief scientists from every major lab — which means they're being treated as credible senior voices on the industry's most consequential policy question. That's not the profile of a company in trouble.

The backlash is what strong positioning costs. When you publicly advocate for oversight, you irritate the open-source community. When you add restrictive guardrails, you irritate developers who want flexibility. When you move aggressively in enterprise sales, you irritate competitors who'd prefer you stay quietly in the safety lane. None of those criticisms prove the strategy is wrong — they might just mean it's working. The companies that draw the most fire are usually the ones gaining the most ground.

The $1 Billion Deal Nobody Is Talking About Enough

While the AI industry debated open weights and guardrails, Cyera — a data-security company — acquired Oasis Security for approximately $1 billion. If you've been following the OpenAI Hugging Face breach, the connection is immediate and jarring. Oasis Security specializes in non-human identity management: controlling and securing the credentials and access rights of automated systems, including AI agents.

The breach that dominated this week's headlines succeeded because an AI agent found credentials from four separate accounts exposed online and chained them into a working cyberattack across multiple services over nine days. Oasis Security exists precisely to manage and protect those kinds of non-human credentials. Cyera paying $1 billion for that capability — in the same week as the breach — is the market's verdict on how serious the AI agent security problem has become. This is Cyera's third acquisition this year, and AI-security acquisitions overall have tripled in 2026. The category is forming fast and the money is very serious.

What AI Agent Security Actually Means for Builders

Here's what's happening at speed across the industry: AI agents are being deployed everywhere before the security infrastructure to protect them exists. Non-human identity management — controlling what AI agents can access, monitoring their behavior in real time, revoking permissions when something goes wrong — is the critical missing piece, and most organizations don't have it in place. Companies like Oasis Security that built in this space early are now worth billions because the demand went from theoretical to undeniable in the span of one incident.

For anyone building with AI agents right now, the practical takeaways are straightforward. Treat your AI agents like privileged user accounts: scope their permissions to the minimum they actually need, rotate the credentials they use, and monitor their behavior actively with real-time alerts rather than passive logs nobody reads. The four-account breach succeeded because ordinary credential hygiene wasn't in place. That's a fixable problem — but only if you treat it as one rather than assuming your agents will behave.

The Closed Labs and the Security Alliance Gap

There's a striking contradiction worth naming directly. The same week that employees at OpenAI, Anthropic, and Google signed a letter asking for government-coordinated AI slowdown infrastructure, those same companies were conspicuously absent from the Open Secure AI Alliance — the Nvidia-led coalition of 30+ companies including Microsoft, IBM, SpaceX, Hugging Face, and the Linux Foundation that formed this week to build shared AI cyber-defense tools.

The distinction is real: the pacing letter was signed by individual employees expressing personal concern, while joining the alliance is a corporate decision with competitive implications around open-source sharing. But the gap still illustrates a pattern: the closed labs support governance in the abstract and future tense while opting out of concrete, present-tense collective action that touches their competitive positions. Asking government to build slowdown tools for a future threat is easier than joining an open alliance to defend against a current one.

My Take

The Anthropic backlash will fade. Strong positioning always draws fire, and the companies that keep winning usually keep drawing it. The Cyera-Oasis deal is the story I'll be thinking about in five years, because it marks the moment AI agent security became a mainstream enterprise priority rather than a niche concern. The organizations that get ahead of this — scoping agent permissions, managing non-human identity, monitoring agent behavior actively — will be far better positioned than those that discover the problem via their own incident. The market just made that math very clear with a billion-dollar acquisition.

What's your experience? Drop a comment below! 👇 Are you more concerned about the Anthropic safety debate or the growing AI agent security threat — and what steps are you taking in your own projects?