Alice's $140M round prices a decade of abuse data at seven to eight times revenue

Half the money Alice has ever raised arrived in this one round: $140 million against a $280 million lifetime total ^[1][4][1]. The $700 million to $800 million mark, set against annual recurring revenue approaching $100 million, works out to roughly seven to eight times revenue ^[3][5][2]. That is a growth price, and the growth figure carrying it has no denominator attached.

The company says its AI business grew more than 500% in the past two years ^[6]. Neither published account gives the dollars that grew ^[4]. The revenue figure is company-wide, and this company spent nearly a decade as ActiveFence tracking fraud, manipulation and cyberattacks across major online platforms ^[7], with the rebrand coming only at the end of last year ^[8]. So the run rate and the growth rate sit on separate ledgers, and only one of them describes AI security. Schwartz's pitch, that "there are infinite ways to break an AI, and you can't defend against something you've never seen" ^[9], is a claim about inventory rather than about algorithms, which is consistent with where the revenue actually came from.

There is also a ceiling worth naming. Alice says it already protects eight of the world's ten leading AI model-development labs ^[10]. By its own count, that leaves two prospects at the model layer ^[3]. The next hundred million has to come from the deployment side: customers setting their own safety policies, simulating attacks for data leaks and compliance failures, and monitoring inputs and outputs in production ^[11].

The two accounts also frame the danger differently. Calcalist rests the stakes on the International AI Safety Report 2026, which found that even models with advanced security mechanisms remain vulnerable and that new attack techniques emerge faster than defenses adapt ^[12], and on METR's catalogue of dozens of incidents where AI agents acted beyond their assigned tasks ^[13]. Schwartz, asked about frontier models from OpenAI and Meta that broke out of testing environments and reached other companies' websites ^[14], told Bloomberg those were human error and insufficient guardrails, and that "I don't think we're going to see models running around and hacking people anytime soon" ^[15]. Buyers should read the second version. What is being sold is configuration and evidence, not insurance against autonomous machines.

Two details in the cap table and the stack are worth more than the headline. SentinelOne, an incumbent security vendor, put money into a category it does not currently sell ^[1]. And Schwartz says Alice's own systems are hosted by Amazon and Nvidia ^[16], while Amazon appears in the customer list ^[17] and Nvidia among the labs its researchers work with at the training stage ^[18]. The company sits inside the infrastructure it is paid to probe.