What are the top AI safety startups by revenue today?

The ranking of top startups in the AI safety market by revenue

Below is a table ranking all the companies in this market by their current revenue scale. You can find our methodology at the end of this page.

If you want a deeper understanding of the market and its current dynamics, get our report covering the AI Safety Market.

What are the top AI safety startups by revenue today?

HiddenLayer is the clearest revenue leader among independent AI safety and AI security startups today.

HiddenLayer CEO Chris Sestito recently told TechCrunch that annual recurring revenue is now in the “tens of millions” of dollars. HiddenLayer’s own funding announcement adds two useful pieces of context: ARR grew more than 10x over the previous year, and the company signed more than 50 new platform customers across finance, technology, government, pharmaceuticals, defense and other industries.

We do not have anything nearly as clean for the companies behind it. Noma Security says ARR grew more than 1,300% in a year but does not disclose the resulting ARR. Patronus AI says revenue grew more than 15x but also keeps the dollar figure private. Credo AI says revenue doubled during 2025. WitnessAI and Zenity have attracted large enterprise customers and substantial funding without publishing current revenue.

So there is a clear number one, followed by a much less certain group.

Current position	Company	Strongest current commercial evidence	What we actually know
1	HiddenLayer	ARR in the tens of millions; ARR grew >10x	Strongest absolute revenue evidence
Next tier	Noma Security	ARR grew >1,300%	Absolute ARR undisclosed
Next tier	Patronus AI	Revenue grew >15x	Absolute revenue undisclosed
Next tier	Credo AI	Revenue doubled in 2025	Absolute revenue undisclosed
Next tier	WitnessAI	Large enterprise deployments; $58M strategic funding round	Revenue undisclosed
Next tier	Zenity	Large enterprise customer base; $125M Series C	Revenue undisclosed
Earlier-stage	Mindgard	Growing enterprise AI security business; $30M Series A	Revenue undisclosed

Which AI safety startup makes the most revenue today?

HiddenLayer is the clearest revenue leader among independent AI safety and AI security startups today.

HiddenLayer CEO Chris Sestito recently told TechCrunch that annual recurring revenue is now in the “tens of millions” of dollars. HiddenLayer’s own funding announcement adds two useful pieces of context: ARR grew more than 10x over the previous year, and the company signed more than 50 new platform customers across finance, technology, government, pharmaceuticals, defense and other industries.

We do not have anything nearly as clean for the companies behind it. Noma Security says ARR grew more than 1,300% in a year but does not disclose the resulting ARR. Patronus AI says revenue grew more than 15x but also keeps the dollar figure private. Credo AI says revenue doubled during 2025. WitnessAI and Zenity have attracted large enterprise customers and substantial funding without publishing current revenue.

So there is a clear number one, followed by a much less certain group.

Does Anthropic count as an AI safety startup in this revenue ranking?

Anthropic should stay outside this AI safety startup ranking because customers mainly pay Anthropic for Claude models and AI products rather than an independent safety product.

Anthropic was founded with AI safety at the center of its identity, so leaving it out can initially look strange. The problem is economic rather than philosophical. Including Anthropic would make model revenue dominate a ranking that is supposed to tell us how large the commercial market for AI security, evaluation, governance and safety infrastructure has become.

The same logic applies to frontier research organizations whose main business is building models. We focus here on companies selling products that help customers test AI, govern AI, secure models and agents, catch dangerous behavior or reduce AI-specific risks.

That definition captures companies such as HiddenLayer, Noma Security, Patronus AI, Credo AI, WitnessAI and Zenity much more cleanly.

How much revenue does HiddenLayer make now?

HiddenLayer currently has annual recurring revenue in the tens of millions of dollars, making it the strongest disclosed revenue business in this group.

The exact ARR remains private. Chris Sestito gave TechCrunch the range rather than a precise number, while HiddenLayer’s own announcement confirmed that ARR had increased more than tenfold over the previous year.

The customer expansion makes that growth more convincing. HiddenLayer says more than 50 new platform customers signed during the period, and more than 90% of its ARR growth came from new customers. The company now sells into securities brokerage, banking, insurance, government, defense, pharmaceuticals, technology and other large enterprise markets.

An older report from The Information gives us a useful baseline. HiddenLayer had only recently crossed roughly $10 million in annualized revenue at that point. The latest disclosure therefore reflects a real step up in scale rather than a minor improvement around the same level.

Is Noma Security already as big as HiddenLayer?

Noma Security could be close to HiddenLayer, but the public numbers still do not prove it.

Noma Security says ARR grew more than 1,300% in a year and that dozens of enterprise customers now use its platform across financial services, life sciences, retail and big technology companies. The company also raised a $100 million Series B after emerging from stealth less than a year earlier.

The missing piece is the starting base. A 1,300% increase can produce very different outcomes depending on where revenue began. Without an absolute ARR figure, we cannot tell whether Noma went from hundreds of thousands to several million dollars or from a few million to several tens of millions.

That puts Noma among the strongest challengers. Putting it ahead of HiddenLayer today would require a number Noma has not disclosed.

How big is Patronus AI now?

Patronus AI is growing extremely fast, although its current revenue remains private.

Patronus AI says revenue grew more than 15x over the past year. The company has also moved beyond basic LLM evaluation into simulation environments where AI agents can be trained and stress-tested across complicated digital workflows.

That shift appears to have widened the customer base. Patronus says it works with leading frontier AI labs and enterprises, while investors involved in its recent $50 million Series B described strong demand for its environments.

The commercial trajectory looks stronger than the older third-party revenue estimates that still circulate online. We would rather use Patronus AI’s direct 15x growth disclosure than pretend one of those stale estimates represents the current business.

Patronus belongs near the front of the market. Its exact position behind HiddenLayer is still impossible to pin down.

How much revenue does Credo AI make today?

Credo AI has become a meaningful enterprise AI governance business, but it still does not publish an absolute revenue figure.

The company’s 2025 review says revenue doubled, enterprise customer count rose 150%, its European business doubled and advisory engagements increased fivefold. Earlier in the growth cycle, TIME had reported that Credo AI’s revenue had quadrupled over the preceding year and that its customer base had doubled.

Those disclosures refer to different measurement periods, so they should not be treated as contradictory versions of the same number. Together, they show several periods of rapid expansion.

Credo AI also sells into recognizable large enterprises, with Mastercard and Cisco among the customers it has publicly discussed.

The evidence is strong enough to place Credo among the larger independent AI safety companies. It is still weaker for a revenue ranking than HiddenLayer’s disclosed ARR range because we do not know the size of Credo’s revenue base.

Are AI governance startups making as much money as AI security startups?

AI security currently shows stronger revenue evidence than AI governance.

The gap is clearest in the disclosures. HiddenLayer has reached an eight-figure ARR range. Noma Security has reported explosive ARR growth. Several acquired AI security companies had also reached meaningful commercial scale before larger cybersecurity vendors bought them.

Governance is growing too. Credo AI’s revenue and customer growth show that large companies are spending real money on AI oversight, compliance and risk management. Regulatory pressure should keep that budget relevant.

Security has an easier route into spending today because CISOs already have established budgets, procurement processes and security platforms. Products that detect unsafe AI use, block prompt attacks, protect agents or stop data leakage can often fit into that existing buying motion.

For now, that gives AI security vendors the clearest route to larger recurring revenue.

How much revenue did the first big AI safety startups make before they were acquired?

Several early AI safety leaders were acquired while their standalone revenue was still surprisingly modest.

The Information reported that Robust Intelligence was generating roughly $9 million in annualized revenue around the time Cisco agreed to acquire the company. The same reporting said Lakera and CalypsoAI had each remained below $5 million in revenue at the point examined.

Those numbers are revealing because all three companies became strategically valuable to major cybersecurity buyers. Cisco integrated Robust Intelligence into its AI security products. Check Point acquired Lakera. F5 acquired CalypsoAI and has since folded its technology into a broader AI security platform.

Protect AI followed a similar path. Palo Alto Networks completed that acquisition and later disclosed $635 million of purchase consideration in its regulatory filings, despite Protect AI never publishing a comparable standalone revenue number.

Prompt Security and Aim Security were also acquired before giving the market much financial transparency.

Startup	Best useful standalone revenue evidence	Buyer	What happened
Robust Intelligence	About $9M annualized revenue	Cisco	Acquired and incorporated into Cisco AI security
Lakera	Below $5M revenue at the period reported	Check Point	Acquired for AI-native runtime security
CalypsoAI	Below $5M revenue at the period reported	F5	Acquired and integrated into F5 AI Security
Protect AI	Revenue not publicly disclosed	Palo Alto Networks	Acquired; $635M purchase consideration later disclosed
Prompt Security	No strong public standalone revenue figure	SentinelOne	Prompt Security ARR later doubled quarter-on-quarter inside SentinelOne
Aim Security	Revenue not publicly disclosed	Cato Networks	Acquired and integrated into Cato AI Security

Why are so many AI safety startups getting acquired so early?

Large cybersecurity companies are buying AI safety startups because specialist technology has become strategically useful before many of these startups have built huge standalone revenue businesses.

Cisco, Palo Alto Networks, Check Point, F5, SentinelOne and Cato Networks have all bought companies focused on securing AI models, applications, prompts or agents.

The economics are straightforward. A specialist startup may have strong technology and a few large customers but still need years to build global enterprise distribution. An established cybersecurity vendor already has thousands of customers, large sales teams and an existing CISO budget.

Prompt Security gives us a good example of what can happen next. After SentinelOne bought the company, SentinelOne reported that Prompt Security ARR was doubling quarter-on-quarter. More recently, SentinelOne said ARR across its AI offerings had nearly tripled year over year.

Acquisitions can accelerate the commercial scale of these products much faster than independent expansion would.

Is WitnessAI already one of the biggest AI safety companies?

WitnessAI has enough enterprise traction to be taken seriously, although there is still no public revenue figure that puts it near the top of the ranking.

WitnessAI raised $58 million in strategic funding and said the money would support international growth and a broader push into AI agent security. Its product gives enterprises visibility and control over employee AI use, AI applications and increasingly autonomous agents.

That positioning is commercially attractive because it sits close to existing enterprise security and governance budgets.

Still, the evidence stops short of revenue. A large funding round tells us investors expect WitnessAI to grow; it does not tell us what customers are paying today.

Until WitnessAI publishes ARR, annual revenue or another comparable financial metric, any precise revenue position would be guesswork.

Is Zenity already one of the largest AI safety startups?

Zenity is one of the best-funded independent AI agent security companies, but Zenity has not disclosed enough revenue data to rank it near HiddenLayer with confidence.

Zenity raised a $125 million Series C and says some of the world’s largest enterprises use its platform to secure AI agents. The company focuses on understanding what agents intend to do and controlling their actions before unsafe behavior reaches production systems.

That is a strong position in one of the fastest-growing parts of the market. Zenity also has much more capital available than most younger AI safety startups.

What we still lack is an absolute commercial figure. Funding, customers and product relevance make Zenity important; they do not tell us whether annual revenue is $5 million, $20 million or substantially more.

Zenity therefore belongs high on the watchlist rather than in a precisely numbered revenue position.

Which newer AI safety startups are starting to matter commercially?

Mindgard is one of several younger AI safety companies worth watching as enterprise spending moves toward continuous AI testing and runtime protection.

Mindgard recently raised a $30 million Series A and says its research has uncovered more than 150 publicly disclosed high-impact vulnerabilities in widely used AI products. The company sells security testing and protection for models, applications and agents.

Lasso Security, Vijil, Straiker and other specialists are attacking adjacent parts of the same problem. Some focus on AI red teaming, some on data leakage, some on runtime enforcement and others on agent permissions or model vulnerabilities.

Most of these companies have not published strong revenue numbers yet. That makes them relevant to the future market without forcing them prematurely into today’s top revenue tier.

The candidate pool behind the current leaders is getting much deeper.

Why is AI agent security becoming such a big business?

AI agent security is attracting money because AI agents can take actions inside real company systems.

A chatbot producing a bad answer is one kind of risk. An agent with access to databases, files, credentials, SaaS tools and code repositories can create a much more expensive failure.

Noma Security now markets directly around agent security. HiddenLayer is expanding its runtime protection for autonomous coding agents. Zenity has centered its recent strategy on controlling agent actions. WitnessAI is extending governance into enterprise agents. Cato incorporated Aim Security into its broader AI security platform.

The product is also easier for enterprise buyers to understand now. Companies increasingly need to answer concrete questions: which agents exist, what those agents can access, what tools they can call, whether an instruction has been manipulated and whether an unsafe action should be blocked.

Those are operational security problems with budgets attached to them.

Can we trust online revenue estimates for AI safety startups?

Third-party revenue estimates are useful as rough context, but they are too inconsistent to drive this ranking.

Private AI safety companies rarely publish audited accounts with detailed revenue, so commercial databases often fill the gap with modeled estimates. The problem becomes obvious when several databases produce materially different numbers for the same startup.

Direct disclosures also age quickly in this market. An estimate based on a startup’s 2024 or early-2025 size can become misleading when the company later reports a tenfold or fifteenfold growth rate.

We therefore use third-party estimates mainly to test whether a number looks plausible or to understand older scale. When a company or a strong primary source provides ARR, revenue or a clear commercial metric, that evidence takes priority.

That approach produces a less tidy ranking, but the uncertainty is real.

Does a huge funding round mean an AI safety startup already has huge revenue?

A huge funding round can point to commercial momentum, but funding remains a weak substitute for actual revenue.

The current market makes that obvious. HiddenLayer raised $100 million, Noma Security raised $100 million, Patronus AI raised $50 million, WitnessAI raised $58 million and Zenity raised $125 million.

The quality of those signals differs. HiddenLayer combined its raise with a direct ARR range. Noma and Patronus disclosed extraordinary growth rates without absolute revenue. WitnessAI and Zenity emphasized enterprise adoption but kept revenue private.

Investors are financing what they expect these markets to become, not simply paying a multiple on today’s revenue.

For revenue ranking purposes, the funding amounts therefore sit behind the financial disclosures rather than replacing them.

How concentrated is AI safety startup revenue right now?

The commercial AI safety market is still small enough that one startup with ARR in the tens of millions stands out clearly.

As seen above, HiddenLayer is the only independent specialist in our research with a current company-backed disclosure that places ARR comfortably in an eight-figure range. Several challengers may also be sizable, but their disclosures give growth rather than absolute dollars.

Historical acquisition data reinforce the point. Some of the best-known first-generation AI security startups were still around or below the $5–10 million revenue range when acquisition discussions began.

The sector therefore has a lot of enterprise activity without yet showing the revenue concentration we see in mature cybersecurity categories.

Commercial scale	Companies	Evidence available
Eight-figure ARR clearly disclosed	HiddenLayer	Current ARR range disclosed by CEO
Potentially large, exact figure unknown	Noma Security	Very fast ARR growth, absolute ARR private
Potentially large, exact figure unknown	Patronus AI	Very fast revenue growth, absolute revenue private
Established enterprise business, exact figure unknown	Credo AI	Revenue and customer growth disclosed
Enterprise traction, revenue private	WitnessAI	Funding and customer adoption
Enterprise traction, revenue private	Zenity	Funding and large-enterprise adoption
Earlier commercial stage	Mindgard and several newer specialists	Product and funding evidence stronger than revenue evidence

Are any pure-play AI safety startups above $100 million in revenue yet?

We found no defensible public evidence that an independent pure-play AI safety startup has already crossed $100 million in annual revenue or ARR.

That threshold matters because many adjacent AI companies have blown through $100 million quickly. Pure-play safety and security specialists are still operating on a smaller commercial base.

HiddenLayer is the closest independent company for which we have strong enough current evidence to discuss the question seriously. Noma Security and Patronus AI could also be much larger than their last known bases suggest, but neither gives us the absolute number needed to claim a $100 million business.

The acquisition pattern may also keep the independent ceiling lower. Large cybersecurity platforms have repeatedly bought promising AI security companies before those companies reached mature standalone scale.

A $100 million pure-play leader now looks plausible. We just cannot point to one yet.

Are the biggest AI safety startups really becoming cybersecurity companies?

Commercially, the AI safety market is increasingly being shaped by cybersecurity products.

The companies with the clearest routes to revenue sell protection around AI applications, models and agents: runtime security, attack detection, prompt-injection defense, asset discovery, agent permissions, red teaming and data-loss controls.

That does not make evaluation or governance unimportant. Patronus AI shows strong demand for agent testing and simulation, while Credo AI has built a growing enterprise governance business.

Security simply has the strongest buying infrastructure today. CISOs already buy software to discover assets, enforce policies, stop attacks and investigate incidents. AI-specific products can attach themselves to those familiar jobs.

This commercial pull explains why traditional cybersecurity vendors have been so aggressive with acquisitions.

Which AI safety startups could challenge HiddenLayer next?

Noma Security and Patronus AI currently have the strongest public growth evidence among the companies that could challenge HiddenLayer.

Noma has broad Fortune 500 adoption and a product spanning AI discovery, posture management, red teaming and runtime security. Patronus has become deeply involved in evaluation and simulation infrastructure used by frontier AI developers.

Zenity is another serious candidate because agent security is becoming a large enterprise problem and the company now has considerable capital to expand. WitnessAI is pursuing a similarly large enterprise opportunity around visibility, governance and agent security.

Credo AI has a different route through governance. Its growth already shows that companies will pay for operational AI oversight, especially as legal and compliance requirements become harder to manage manually.

The order behind HiddenLayer could move quickly because several of these businesses are growing from relatively small bases.

What are the top AI safety startups by revenue today?

HiddenLayer is the clearest number one by current disclosed revenue evidence, while Noma Security and Patronus AI form the strongest challenger group.

The gap between “largest company” and “best disclosed number” is important here. Some private competitors may already have more revenue than the public evidence shows. We cannot rank hidden revenue.

Credo AI has convincing enterprise growth but no absolute number. WitnessAI and Zenity have substantial enterprise backing without published revenue. Younger companies such as Mindgard are building quickly but remain earlier in the commercial cycle.

The broader finding is more interesting than a forced top-ten ordering. AI safety has moved from a market where leading specialists often had only a few million dollars of revenue into one where the strongest independent company has reached a much higher commercial tier and several challengers are growing at unusually fast rates.

For now, HiddenLayer is the company we can put at the top with the least hesitation. The race behind it remains genuinely open.