Methodology
Primary sources, one link per number, a date on every row.
Where the data comes from
We read each vendor’s own pricing pages, documentation, status pages, filings and press releases, plus named industry reports. Funding rounds come from the company’s own announcement or named trade press (SecurityWeek, SiliconANGLE, Business Wire and similar); deal closing dates come from the buyer’s release or SEC filing. Every record lists its sources with the date we accessed them, and every number in the table links to the source it came from.
How we verify
Each row shows the date we last checked it. A field is filled only when a source states it; when a vendor does not publish a figure, the cell stays empty. List prices are the public, on-demand prices; negotiated and committed-use discounts are not shown.
What a Yes means
A Yes in a capability column means the vendor documents that capability on its own product pages, docs or press releases. A dash means no source we read says so; it does not mean the product lacks the feature. Each Yes carries the sentence the vendor published, on the vendor page, with its source link and the date we read it.
Which vendors are listed
One row per buyer-facing AI security product line. When a platform vendor bought a startup, the startup's capabilities appear under the buyer's row and the deal appears in the acquisition ledger, as the parties state it. A deal that has been announced but not completed is marked pending. Vendors with fewer than three sourced fields are left out.
Vendor claims
Latency, attack-library size and detector counts are figures the vendor publishes. We record them as claims and have not benchmarked them. Framework columns list the frameworks a vendor says its controls or reports map to; a mapping is not a certification.
Corrections
Vendors and readers can send corrections to hello@aisecurityvendors.com with a public source. We update the row and its verified date. Featured placement is paid and labelled; it never changes a row’s data or its position in the sorted table.
Reuse
The dataset is free to reuse under CC BY 4.0 with a link to aisecurityvendors.com. Download it as CSV or JSON.
Field definitions
22 fields- Runtime guardrails
- Inspects prompts and model responses at runtime and blocks or redacts attacks such as prompt injection and jailbreaks. Yes means the vendor documents it on its own pages; a dash means no source says so.
- Agent security
- Controls or tests AI agents and their tools, including MCP servers and tool calls. Yes means the vendor documents it on its own pages; a dash means no source says so.
- Red teaming
- Runs adversarial tests against models, applications or agents before or after deployment. Yes means the vendor documents it on its own pages; a dash means no source says so.
- AI-SPM and AI-BOM
- Discovers and inventories AI assets (models, agents, apps, shadow AI) and assesses posture, misconfigurations or AI-BOM. Yes means the vendor documents it on its own pages; a dash means no source says so.
- Model and supply-chain scanning
- Scans models, model files or the AI and agent supply chain for tampering and malicious code. Yes means the vendor documents it on its own pages; a dash means no source says so.
- Data protection for AI
- Detects, redacts or governs sensitive data in prompts, responses, training data and AI workflows. Yes means the vendor documents it on its own pages; a dash means no source says so.
- Deal status
- Whether the vendor has bought an AI security company, has a deal pending, or is now owned by a larger company, per press releases and vendor pages.
- Self-host or private cloud
- The vendor documents running the product on your own servers, in your own cloud or VPC, or air-gapped. Derived from the deployment text on the same source.
- Ownership and deals
- The acquisition or ownership fact, as stated by the parties.
- Named modules
- Named AI security modules the vendor lists.
- Frameworks mapped
- Security and compliance frameworks the vendor explicitly maps its testing, controls or reports to. A mapping, not a certification.
- Integrations and platforms
- Named clouds, models, development tools, protocols or security tools the product supports.
- Deployment
- Documented hosting and installation options.
- Licensing
- Open-source status or paid packaging, where documented.
- Content inspected
- Content formats the product inspects, where documented.
- Vendor latency claim ms
- Upper bound of the runtime latency the vendor claims ("below N ms"). Vendor-published, not independently benchmarked.
- Attack library attacks
- Lower bound of the attack simulations the vendor says it runs ("more than N"). Vendor-published.
- Sensitive data types detected types
- Count of PII and confidential-data types the vendor says its detector covers.
- Languages monitored languages
- Lower bound of languages the vendor says it monitors in real time.
- AI services covered services
- Lower bound of the AI services the vendor says its discovery covers.
- Latest disclosed round $M
- Size of the most recent funding round the company or the press reported, in US$ millions.
- Funding
- Latest round, lead investor and month, plus the total raised where the source states it.