What’s funding like in the AI safety market right now?

All funding deals in the AI safety market over the last 24 months

Below is the table listing all the deals. 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.

Is AI safety funding actually healthier than it was a year ago?

AI safety funding looks healthier today: more companies are raising, the typical round is bigger, and the market has developed a much deeper growth-financing layer.

We counted 76 deals between September 20, 2025 and September 19, 2026, up from 61 during the previous 12 months. The number of unique funded companies rose from 59 to 71, while median round size went from $6.75 million to $9.25 million.

If capital had doubled while the number of companies fell, we would be looking at a narrow market dominated by a few winners. That is not what happened here: company breadth increased by 20.3% while deal activity grew 24.6%.

New-company formation also held up. First financings increased from 29 to 37, and the number of identifiable lead or strategic investors we tracked rose from 82 to 102.

The part that has changed most is how far successful companies can now go. Later-stage rounds gained share, Series A and B checks became much larger, and follow-ons absorbed far more capital. AI safety today looks like a young market that has started producing a real group of scale-ups.

Metric	Latest 12 months	Previous 12 months	Change
Deals	76	61	+24.6%
Unique companies	71	59	+20.3%
Total capital	$1.690B	$853.5M	+98.0%
Median round	$9.25M	$6.75M	+37.0%
First financings	37	29	+27.6%
Follow-ons	38	29	+31.0%
Captured lead/strategic investors	102	82	+24.4%
Countries represented	12	9	+33.3%
Later-stage deal share	10.5%	6.6%	+3.9 pp

Does AI safety funding still look strong without the biggest rounds?

Yes. Even after stripping out the largest AI safety financings, the market still grew much faster than deal count.

Total capital rose 98%, from $853.5 million to $1.69 billion. But removing every round above $50 million still leaves $793.5 million of current funding, up 57.6% from $503.5 million previously. Excluding rounds above $100 million gives a similar 53.7% increase.

The monthly numbers also look stronger than they did a year ago. Median monthly funding climbed from $69.3 million to $126.3 million, an increase of roughly 82%. Neither period had a zero-deal month.

We can see the shift further down the round-size distribution too. Deals below $5 million fell from 41.0% of transactions to 30.3%, while the $20 million to $50 million bracket jumped from 8.2% to 18.4%. The $50 million to $100 million bracket rose from 3.3% to 9.2%.

So the recent $100 million-plus rounds are sitting on top of a broader increase in normal financing sizes. They are amplifying the boom rather than creating the whole thing.

Are more AI safety startups getting funded now?

Yes. AI safety funding is reaching more startups now, including more companies raising for the first time.

Unique funded companies increased from 59 to 71, and first financings rose from 29 to 37. First rounds give us a rough view of whether investors are still feeding new companies into the market rather than concentrating entirely on businesses they already know.

The expansion also crossed more borders. We found companies headquartered in roughly 12 countries during the latest period compared with nine previously.

There were more repeat fundraisers at the same time. Five companies completed multiple financings during the latest 12 months versus only one in the previous window. Even with those additional repeat rounds, unique-company growth stayed close to the growth in total deals.

AI safety is therefore adding new funded startups while also producing more companies capable of raising repeatedly.

Is AI safety funding getting concentrated in a few winners?

No. Large AI safety rounds have become much more common, but the market is actually less dependent on its top few deals.

The three largest financings accounted for 22.4% of current capital, down from 32.8% in the previous period. The top-five share fell from 46.3% to 33.1%, and the largest single financing represented 8.9% of the market compared with 11.7% previously.

At the same time, we counted ten rounds above $50 million versus four a year earlier. Three current deals exceeded $100 million.

Goodfire raised $150 million, Zenity's Series C contained roughly $115 million of primary capital, Onyx Security raised $113 million and HiddenLayer raised $100 million. Adaptive Security, Braintrust, Neo, Straiker, Arcade.dev and WitnessAI also landed rounds between $58 million and $81 million.

A much larger group of companies can now raise serious growth capital, rather than the market being carried by one or two giant rounds.

Are investors still backing new AI safety startups, or mostly funding follow-ons?

Investors are still backing new AI safety startups, but the big money has swung hard toward companies that have already raised before.

First financings increased from 29 to 37, and their median size rose from $3.9 million to $5.1 million. The entry point into the market remains active.

Follow-ons changed much more dramatically. We counted 38 versus 29 previously, while the median follow-on jumped from $10.9 million to $29 million. Total follow-on capital reached about $1.395 billion.

Among deals where we could establish financing status, first and follow-on rounds were almost evenly split by count. By dollars, however, follow-ons took roughly 83% of current capital compared with about 68% previously.

Investors are still creating new funded companies. Once a company starts showing enough progress to come back for another round, the amount of money available is now on a completely different scale.

Measure	Latest 12 months	Previous 12 months
First financings	37	29
Median first financing	$5.1M	$3.9M
Capital in first financings	$293.8M	$240.5M
First-financing share of known-status capital	~17%	~32%
Follow-ons	38	29
Median follow-on	$29M	$10.9M
Capital in follow-ons	$1.395B	$521.7M
Follow-on share of known-status capital	~83%	~68%

Where is AI safety funding moving right now?

Right now, AI safety money is moving most clearly toward guardrails, agent control and other products that sit close to AI systems running in production.

AI Guardrail Platforms went from ten financings to 31. Their share of all deals rose from 16.4% to 40.8%, while their share of capital jumped from 11.8% to 42.9%.

AI Safety Monitoring also attracted much more money without a comparable jump in deal count. Capital rose from $65.5 million to $195.6 million while its share of deals stayed around 14.5%.

Model Robustness Tools look impressive in headline dollars, but the picture there is much narrower. Goodfire's $150 million Series B accounts for most of the increase.

AI Evaluation Tools went in the opposite direction, with deal count falling from 14 to seven and total capital dropping from $310.1 million to $177.9 million.

The funding rotation is quite clear now: investors are putting much more money into controlling and monitoring deployed AI systems, especially once those systems can act autonomously.

Category	Current deals	Capital	Deal share	Capital share	What changed
AI Guardrail Platforms	31	$724.1M	40.8%	42.9%	Broad surge
AI Risk Platforms	16	$314.0M	21.1%	18.6%	More capital, lower share
AI Safety Monitoring	11	$195.6M	14.5%	11.6%	Bigger rounds
Model Robustness Tools	5	$188.4M	6.6%	11.1%	Goodfire-driven
AI Evaluation Tools	7	$177.9M	9.2%	10.5%	Fewer deals
AI Red Teaming	6	$89.6M	7.9%	5.3%	Losing share
Total	76	$1.690B	100%	100%	Broader market

Are AI safety investors moving away from evaluation tools?

Yes, AI evaluation funding has cooled sharply even though the companies still getting funded are raising larger rounds.

AI Evaluation Tools dropped from 14 deals to seven. Capital fell about 43%, from $310.1 million to $177.9 million, while the category's share of total funding went from 36.3% to 10.5%.

Yet its median round increased from $6 million to $10 million. Braintrust is a good example of what the surviving end of the category looks like: it raised an $80 million Series B and described its business around the infrastructure needed to evaluate and observe AI once companies move from experiments into production.

We read this as a tougher market for standalone evaluation startups rather than investors abandoning evaluation itself. Evaluation is increasingly becoming part of broader production, observability and security stacks.

The money has moved fastest toward products that can control what AI does after deployment.

Is AI safety finally moving beyond Seed rounds?

Yes. Seed still produces most AI safety deals, but Series A and Series B now control a much larger part of the money.

Seed represented 65.8% of current transactions, down from 75.4%. Series A moved from 14.8% to 21.1% of deals, while Series B rose from 4.9% to 7.9%.

The difference in capital is much bigger. Series A now accounts for 30.1% of funding and Series B for 34.0%. Together they attract nearly two-thirds of the dollars in the market.

Round sizes have moved up with them. Median Series A increased from $18 million to $30 million, while median Series B rose from $45 million to $90.5 million.

Seed still matters enormously for company formation, but these days it no longer tells the whole story of AI safety venture funding. A meaningful growth-stage market now sits above it.

Stage / grouping	Current deal share	Previous deal share	Current capital share	Previous capital share	Current median
Seed	65.8%	75.4%	23.8%	39.3%	$5.45M
Series A	21.1%	14.8%	30.1%	20.4%	$30M
Series B	7.9%	4.9%	34.0%	19.2%	$90.5M
Series C	2.6%	1.6%	8.6%	8.2%	$72.5M
Unknown	2.6%	3.3%	3.5%	12.9%	$29.7M
Early stage	86.8%	90.2%	53.9%	59.7%	—
Later stage	10.5%	6.6%	42.6%	27.4%	—

Why are AI agent security startups raising so much money now?

AI agent security startups are raising so much money because enterprises increasingly need to control what autonomous AI can access and do, not simply check the quality of its answers.

That shift shows up in several recent rounds. HiddenLayer's $100 million Series B is funding deeper agentic runtime protection and security for autonomous coding agents. WitnessAI raised $58 million while expanding tools for discovering and governing enterprise agents. Straiker's $64 million Series A came after the company reported working with Fortune 500 businesses and frontier AI labs.

Onyx Security gives us an especially recent example. When it announced its $113 million Series B, the company said revenue had quadrupled in the four months since it came out of stealth. It also reported securing more than 1.1 million agents and analyzing tens of millions of AI sessions in real time.

Zenity's latest financing tells a similar story from another angle. The company describes autonomous agents spreading through finance, customer service, engineering, HR and security rather than remaining inside isolated AI experiments.

As of now, the security problem is increasingly about permissions, tools, data access and runtime actions. That is a much broader enterprise problem than testing whether a model gives a bad answer.

Are AI safety investors getting pickier?

AI safety investors look pickier about who gets a large check, even though the investor pool itself has become broader.

We tracked 102 identifiable lead or strategic investors in the latest period, up from 82. There is little sign of investors collectively walking away.

What changed is repeat conviction. Eight investors appeared more than once compared with three previously. Andreessen Horowitz appeared in four captured financings, SYN Ventures in three, while Norwest, Redpoint, Bessemer Venture Partners, Hetz Ventures, LG Technology Ventures and Ugly Duckling Ventures each appeared twice.

The size gap between first rounds and follow-ons reinforces that picture. Early companies can still get funded, but investors are committing much more capital once a company shows enough progress to raise again.

There is more competition for deals across the market while the biggest checks remain concentrated around companies that have already built some credibility.

Are corporate investors taking a bigger role in AI safety?

Yes. Corporate and strategic investors are showing up more often in AI safety deals, and they are increasingly joining larger rounds.

Strategic-backed financings increased from 12 to 18. Their participation rate rose from 19.7% to 23.7%, while the median strategic-backed round climbed from $16.25 million to $37 million.

WitnessAI's $58 million round included Samsung Ventures and Qualcomm Ventures. HiddenLayer's $100 million Series B included Microsoft's M12 and Booz Allen Ventures. Arcade.dev's $60 million Series A included strategic investment from Morgan Stanley and Wipro.

Adaptive Security also brought in NVIDIA, Capital One Ventures and Citi Ventures alongside Bain Capital Ventures in its $81 million Series B. The company said at the time that it was already protecting more than 500 enterprise customers.

Corporate investors appear to be following AI safety deeper into actual enterprise deployment. Their role is moving beyond small experimental bets and into companies building infrastructure that sits close to their own technology or customers.

Is AI safety funding becoming more US-centric?

Currently, the biggest AI safety checks are becoming more US-centric even though the overall market is spreading across more countries.

North America's share of deals slipped from 65.6% to 61.8%, while its share of capital jumped from 64.5% to 84.4%. Median financing in the region more than doubled from $7 million to $15 million.

Europe went the other way. Its deal share climbed from 14.8% to 23.7%, but its median round fell from $4 million to roughly $2.84 million.

That suggests Europe is contributing more companies at the early end of the market, while North America is producing most of the large scale-up rounds.

Israel's capital share also fell sharply, from 30.0% to 7.8%, although the earlier period was unusually strong because it contained Noma's $100 million Series B and Irregular's roughly $80 million of funding announced across two closely spaced rounds. Israeli median financing actually increased to $18.5 million in the latest period.

Geographic breadth has improved, but the deepest pools of growth capital are now even more heavily centered in North America.

Are AI safety startups raising again faster?

For now, we do not see convincing evidence that AI safety startups are returning to market faster.

Among current follow-ons where we could reconstruct a previous financing from another public round, the median gap was about 15.1 months. The comparable figure in the previous period was 9.4 months.

We would put very little weight on that apparent slowdown because coverage is uneven: we could reconstruct 15 current follow-ons but only four from the earlier period. Four companies are nowhere near enough for a confident market-wide conclusion.

What we can say more comfortably is that current follow-ons often involve very large jumps in round size. Fourteen of the 15 reconstructed current rounds were more than twice as large as the company's previous known financing, with a median step-up of about 3.05 times.

Goodfire is the extreme example. It went from a $7 million Seed to a $50 million Series A and then a $150 million Series B. That tells us a lot more about the amount of capital available to breakout companies than the weak year-over-year velocity comparison does.

Do AI safety startups need more proof to raise now?

The biggest AI safety rounds today increasingly come with real customer or deployment proof, although we cannot show that every startup now faces a tougher funding bar.

Company age has barely moved. The median funded business was one year old in both periods, and the median follow-on company was two years old. Current Series A companies are typically around two years old and Series B companies around 2.5 years old.

Yet these young businesses are now raising much larger rounds. The market has managed to push companies further up the financing ladder without waiting for them to become much older.

The recent company evidence points toward commercial proof playing a bigger role at the top end. Straiker reported Fortune 500 and frontier-AI customers around its Series A. Adaptive Security said it had more than 500 enterprise customers when it raised its Series B. Onyx reported rapid revenue growth shortly before its $113 million round.

We cannot turn those examples into a clean market-wide percentage because milestones are not consistently disclosed. Still, the pattern around the largest rounds is fairly clear: young companies can raise huge amounts, but the strongest examples now tend to come with evidence that enterprises are already using the product.

Are AI safety companies raising to build products or to scale them?

These days, many of the largest AI safety rounds are funding expansion around products that are already in the market.

WitnessAI said its $58 million financing would accelerate global go-to-market and product expansion. Straiker followed its Series A by hiring a chief revenue officer specifically to scale its international commercial operation. HiddenLayer said its $100 million round would deepen its enterprise platform while expanding runtime protection for agentic systems.

Onyx plans to use its Series B both for new proprietary models and for scaling enterprise sales in the United States and beyond. Zenity's financing also centers on product development and global expansion around customers already deploying autonomous agents.

Arcade.dev offers another useful example. Its $60 million Series A is aimed at building what the company calls the secure action layer for enterprise agents, with Morgan Stanley and Wipro participating strategically.

There are still plenty of early rounds funding product creation and research. But at the upper end of AI safety funding, the language has shifted toward enterprise rollouts, go-to-market expansion, runtime infrastructure and international scale.

Is regulation actually driving AI safety funding?

Regulation is giving AI safety companies another reason for customers to buy, but enterprise AI deployment looks like the bigger funding driver right now.

The EU AI Act became much more concrete during the period we studied. General-purpose AI obligations started applying in 2025, including documentation requirements and, for models with systemic risk, risk assessment, mitigation, serious-incident reporting and cybersecurity duties.

The European Commission's enforcement powers for those GPAI obligations subsequently came into effect, making compliance a more immediate operational issue for affected providers.

That clearly helps companies selling governance, assurance, monitoring and security products. But the deals attracting the most money go well beyond European compliance.

HiddenLayer, Onyx, WitnessAI, Straiker, Zenity and Arcade.dev are all talking about controlling autonomous agents, securing tools, inspecting runtime behavior or governing what AI can do inside an enterprise. Several of the biggest rounds are also going to US companies.

Regulation is reinforcing demand. The faster-moving commercial story is that companies are giving AI access to real systems, real data and real workflows, which creates security problems whether regulators are involved or not.

What does the $1.69 billion AI safety funding headline miss?

The $1.69 billion AI safety funding headline misses the most interesting change: this market now has a much deeper middle and growth layer than it had a year ago.

Looking only at total capital could make the recent period seem like another venture boom built around a few unusually large rounds. The figures underneath tell a different story. Funding below $50 million still grew by more than half, more companies raised, first financings increased and normal round sizes moved up.

The category mix changed too. As seen above, guardrail and runtime-security companies gained broad deal momentum rather than depending on one blockbuster transaction.

Meanwhile, Series A and B became much more important, and follow-on companies captured the majority of new dollars.

The better description of the last 12 months is that AI safety moved further from an early-stage collection of model-risk startups toward a real enterprise-security market with a growing class of scale-ups.

What really changed over the last 12 months?

The biggest change is that AI safety became broader and deeper at the same time. More startups raised capital, more first financings occurred and more investors participated, while the typical financing also became larger.

The second change is where the money goes once a company starts working. Follow-on companies now have access to dramatically larger checks, which has created a visible separation between getting an AI safety startup funded and scaling one.

The third is the move toward agent security and runtime control. Guardrails, monitoring and production security have taken a much larger share of attention as enterprises give AI systems access to tools, code, data and business processes.

The fourth is the arrival of a proper Series A/B market. Seed remains the most common stage, but the financial center of gravity has moved upward. Very young AI safety companies can now reach financing levels that were much rarer one year earlier.

The fifth is that larger rounds increasingly come with signs of real deployment. Customer counts, enterprise contracts, revenue growth and production usage now appear around several of the biggest financings, even though public disclosure is too uneven for a clean market-wide milestone statistic.

The most misleading takeaway would therefore be that AI safety funding simply doubled. The more useful conclusion is that the market has started maturing quickly while keeping its pipeline of new startups alive.