Gray Swan
Enterprise security platform that protects AI agents and models through adversarial evaluation, red‑teaming, and real‑time input/output protection.
- Business Profile
- Founders & Team6 sources scanned
- Market Size9 sources scanned
- Competitor Research5 sources scanned
- Funding6 sources scanned
- Momentum7 sources scanned
- Customer Sentiment7 sources scanned
- AI Visibility
- Risks & Red Flags1 source scanned
Summary
Worth a closer lookGrayswan.ai is an adversarial-first enterprise security platform that protects AI agents and models via red‑teaming, adversarial evaluation and real‑time I/O protections — a deliberate wedge against broader governance/playbooks. Versus Enkrypt AI Guardrails, Akto, White Circle, Noma and Onyx, Grayswan trades breadth for depth: deeper attack simulation and runtime I/O defenses where rivals emphasize automation, Gartner recognition or Fortune‑500 footprint. Recent momentum — a $40M Series A co‑led by Wing and Madrona (June 2026), two $5M early raises (2024/2025) for a disclosed $50M total, a May 28, 2026 Snowflake integration and public Arena challenge outputs with OpenAI/Anthropic engagement — shows the company moving from research to commercial scale. The central tension is whether this technical credibility and platform partnership convert into repeatable enterprise bookings and durable differentiation in a crowded vendor market; first meeting should probe ARR, flagship customers, Snowflake go‑to‑market mechanics and technical moat.
Bull case
- $40M Series A (late May/early June 2026) co‑led by Wing Venture Capital and Madrona with participation from Snowflake Ventures and others, validating investor conviction at scale.
- A platform integration with Snowflake announced May 28, 2026, which can materially accelerate enterprise access and data‑plane workflows if it is a deep technical/GTMS partnership.
- Active public research and community momentum — Arena challenge results and blog releases plus reported collaboration/engagement with OpenAI and Anthropic — that demonstrate technical leadership in adversarial testing.
- Clear product differentiation: emphasis on adversarial red‑teaming and real‑time I/O protection that complements (rather than duplicates) governance incumbents' runtime guardrails.
Watch-outs
- Crowded competitive landscape where Enkrypt has Gartner recognition and automation, and Akto shows stronger Fortune‑500 traction — competitors may out‑scale or out‑sell Grayswan on procurement criteria.
- Mixed external trust/visibility signals (ScamAdviser score ~61; WHOIS privacy, low traffic and registrar flags) and omission from broader vendor lists could indicate limited commercial visibility or immature GTM.
- Disclosed funding totals $50M (Series A $40M + two $5M raises); runway and current ARR/enterprise revenue conversion are not public — commercial traction is not yet independently verifiable.
- Public Arena challenge judging friction (early automated judging issues) signals product/user‑experience rough edges that could matter in enterprise pilots.
Questions for the founder
- 1Current ARR, annual growth rate, and projected runway post‑Series A?
- 2Top 5 customers (names or anonymized verticals), average contract value (ACV) and net retention rate?
- 3Technical depth of the Snowflake integration — is it co‑sell/co‑market, and can you show a live reference or deployment architecture?
- 4How does Grayswan measure and defend its adversarial testing fidelity versus Enkrypt/Akto (IP, datasets, test library provenance, repeatability)?
- 5Sales motion metrics: sales cycle length for enterprise, CAC, CAC payback period, and typical pilot→paid conversion rate?
Company profile
Customers on site
Pricing
Not publicly disclosed. The company website does not list its pricing on any of the 6 pages we read.
Target segments
Product suite
- Shade
Autonomous adversarial red‑teaming solution that runs attack campaigns against AI agents to surface vulnerabilities.
- Cygnal
Runtime protection and monitoring that classifies and blocks adversarial inputs and unsafe outputs in real time.
- Arena
Platform that hosts competitions and a global network of adversarial researchers to discover emerging AI vulnerabilities.
Stats they publish
Social accounts
Founders & Team
Couldn’t confirm the team from public sources for this company.
Market Size
Published market research places the global agentic/cybersecurity-agentic-AI security market in the low-single-digit billions today with multi‑billion forecasts (MarketsandMarkets projects USD 13.52B by 2032; Mordor estimates USD 2.43B in 2026). A defensible SAM for Grayswan is the vendor/software-platform slice of that market (≈ USD 965M in 2026 using Mordor's software-platform share); a conservative 3‑year obtainable scenario is ~2% of that SAM (≈ USD 19.3M).
- MarketsandMarkets ↗the agentic AI security market is projected to grow from USD 1.65 billion in 2026 to USD 13.52 billion by 2032, at a CAGR of 42.0% during the forecast period. · Global · 42% CAGR
- Mordor Intelligence ↗The cybersecurity agentic AI market size is projected to be USD 1.83 billion in 2025, USD 2.43 billion in 2026, and reach USD 9.63 billion by 2031, growing at a CAGR of 31.71% from 2026 to 2031. · Global · 31.71% CAGR
- Grand View Research (Horizon statistics) ↗The global security and surveillance segment generated a revenue of USD 574.9 million in 2025 and is expected to reach USD 14,562.7 million by 2033. The market is expected to grow at a CAGR (2026 - 2033) of 50.4% by 2033. · Global (security & surveillance AI agents segment) · 50.4% CAGR
- BCC Research (AI Agents: Technologies, Applications and Global Markets) ↗is projected to grow from $8 billion in 2025 to $48.3 billion by the end of 2030, at a compound annual growth rate (CAGR) of 43.3% from 2025 to 2030. · Global (AI agents market, broader) · 43.3% CAGR
- Cybersecurity agentic-AI market (2026)USD 2.43 billion (Mordor Intelligence projection for 2026)https://www.mordorintelligence.com/industry-reports/cybersecurity-agentic-artificial-intelligence-market
- Software-platforms share of cybersecurity agentic-AI market39.71% (Mordor Intelligence reported share for software platforms in 2025; applied to 2026 market to estimate platform/vendor SAM)https://www.mordorintelligence.com/industry-reports/cybersecurity-agentic-artificial-intelligence-market
- SOM obtainable share (3 years)2% of the software/platform SAMAssumption based on company stage (growth; 11–50 employees), modest public traction (Arena engages >15,000 adversarial researchers) and a competitive landscape that includes large incumbent security vendors (see MarketsandMarkets key players list)
- No ACV availableGrayswan's ACV or published pricing was not available, so a bottom-up (customers × ACV) SAM was not possible.Target data provided (no pricing captured)
- Published estimates use differing definitions ("agentic AI security", "cybersecurity agentic AI", broader "AI agents") — scope and boundaries vary between vendors; reported CAGR and totals are forecast figures, often paywalled.
- SAM is an estimate derived by applying Mordor's 2025 software-platforms share (39.71%) to Mordor's 2026 market size; that share may shift year-to-year and reports use different segmentation methodologies.
- No company ACV/pricing was available, so a bottom-up customers×ACV SAM was not possible; SAM therefore uses a top-down segment filter rather than Grayswan-specific pricing.
- SOM is a scenario (not a claim) based on a conservative 2% obtainable share in ~3 years; if Grayswan publishes ACV, customer counts, or larger traction metrics, a bottom-up SOM should replace this estimate.
- Multiple credible reports exist with materially different totals (MarketsandMarkets vs Mordor vs Grand View vs BCC); use caution when comparing figures because each uses different definitions and forecast horizons.
Competitors
Top 5Market splits between runtime guardrails/governance platforms and red‑teaming/adversarial specialists. Grayswan positions as adversarial/red‑team and real‑time I/O protection specialist against broader governance/automation incumbents.
- Enkrypt AI Guardrailsenkryptai.com ↗Runtime guardrails for agents and RAG; automated security/compliance. Grayswan competes on deeper adversarial red‑teaming and I/O protection; Enkrypt wins on automation and Gartner recognition.
- Akto Homegrown AI and GenAI Securityakto.io ↗Homegrown AI/GenAI security with continuous red‑teaming and enterprise deployments. Grayswan overlaps on red‑teaming; Akto stronger in Fortune‑500 traction and broad enterprise controls.
- White Circle Control Your AIwhitecircle.ai ↗Automated red‑teaming and low‑latency custom guardrails for agent flows. Grayswan competes on adversarial eval and real‑time I/O protection; White Circle focuses on low‑latency runtime controls.
- Noma Securitynoma.security ↗Discovery, governance and protection across agentic/homegrown AI. Grayswan is more attack‑focused (red‑teaming/adversarial testing); Noma wins on holistic governance and visibility.
- Onyxonyx.com ↗Secure control plane for agent adoption, observability and ROI measurement. Grayswan emphasizes adversarial testing and runtime I/O protection; Onyx wins on adoption governance and metrics.
Funding
- early-stage capital$5M2024Technical.ly ↗
- early-stage capital$5M2025Technical.ly ↗
- Series A$40MJune 1, 2026Wing Venture Capital leadMadrona leadObvious VenturesSnowflake VenturesHudson River TradingSamsung NextMagarac Venture PartnersFinSMEs ↗
Gray Swan publicly announced a $40M Series A in late May/early June 2026 co-led by Wing Venture Capital and Madrona with participation from Obvious Ventures, Snowflake Ventures, Hudson River Trading, Samsung Next and Magarac Venture Partners. A Technical.ly article (citing PitchBook) also reports two earlier $5M early-stage raises in 2024 and 2025, which — if counted — bring the itemized disclosed total to $50M.
- The $40M Series A is widely reported and is corroborated by the company press release and multiple outlets; that round listing and its participants are high-confidence.
- The two earlier $5M rounds are reported only in a Technical.ly article citing PitchBook (the article's sentence attributes those amounts to PitchBook). Those earlier raises are single-source within the provided material and lack independent press releases in the supplied sources.
- Several tracker entries (Tracxn / PitchBook excerpts in the provided text) list total funding as $40M (one round), creating a conflict with the $50M itemized sum when the two $5M raises are included. The itemized $50M is the sum of the rounds documented here; the public record (company and many trackers) commonly reports $40M total — this discrepancy may indicate missing/unclear early-round reporting.
- No valuation is reliably sourced in the cited URLs in the provided text; a $200M valuation phrase appears in an uncited fragment and is not included here.
- Investors listed at the company/press-release level are attributed to the Series A round only when explicitly tied to that round in the cited text; earlier rounds had no round-level investor details in the supplied material.
- Public reporting is largely based on company announcements and press coverage; further confirmation (e.g., SEC filings, detailed PitchBook/Crunchbase entries) may change the itemized history.
Momentum
- FundingJune 1, 2026Gray Swan raised a $40M Series A led by Wing Venture Capital and Madrona.FinSMEs ↗
- Partnership05-28-2026Gray Swan announced a partnership with Snowflake to natively integrate its runtime AI protection into Snowflake’s AI ecosystem.National Law Review ↗
- MilestoneMarch 18, 2026Gray Swan’s Indirect Prompt Injection Arena tested 13 frontier models, 464 red teamers, and 272,000+ attacks; no model came out clean.Gray Swan Arena Blog ↗
- MilestoneSeptember 23, 2025Published results from the Dangerous Reasoning Challenge offering exploits and lessons for AI red-teamers and builders.Gray Swan Arena Blog ↗
- PressSeptember 10, 2025Published an AI security 'Blue Teaming' defense guide covering prompt injection, data poisoning and other attacks.Gray Swan Arena Blog ↗
- PressJuly 28, 2025Released an AI Red‑Team & Prompt Engineering Resource Guide linking 150+ repositories and tools.Gray Swan Arena Blog ↗
- LaunchOctober 17, 2024Launched RepE Chat, an interactive interface based on Representation Engineering (RepE).Gray Swan Arena Blog ↗
- Funding2025Reported prior early-stage raise of $5M in 2025 (per PitchBook citation).Technical.ly ↗
- Funding2024Reported prior early-stage raise of $5M in 2024 (per PitchBook citation).Technical.ly ↗
Dated signals across 2025–2026 show active product/research outputs (Arena challenge results and multiple blog releases), a platform integration with Snowflake (May 28, 2026), and a $40M Series A (June 1, 2026), indicating accelerating commercialization and partnerships alongside continued research. The pattern suggests the company is moving from research toward scaling enterprise adoption.
- Several items come from the company’s own blog or press releases; independent coverage exists for the Series A but much product/research detail is self-published.
- Different outlets report slightly different dates for the Series A (May 28–June 1, 2026); the Series A signal uses the June 1, 2026 FinSMEs date where provided.
- Prior $5M rounds (2024, 2025) are reported via a PitchBook citation inside coverage rather than standalone press releases; details beyond the cited sentence were not provided here.
Customer Sentiment
12- Site flagged as legitimate / not a scamScamAdviser reports the site (app.grayswan.ai / grayswan.ai) as probably legit or safe to use.“It seems that app.grayswan.ai is <strong>legit and safe to use and not a scam website</strong>.”ScamAdviser (app.grayswan.ai) ↗
- ScamAdviser trust score reportedScamAdviser published a numeric review score for the main domain (61), indicating an algorithmic trust assessment.“Our Scamadviser algorithm reviewed grayswan.ai with a score of 61.”ScamAdviser (grayswan.ai) ↗
- Valid SSL / basic technical hygieneScamAdviser notes a valid SSL certificate for the site, listed as a positive highlight across subdomains.“According to the SSL check the certificate is valid”ScamAdviser (grayswan.ai) ↗
- Community engagement via public challengeGray Swan ran a public 'Harmful AI Assistant Challenge' that attracted security professionals and hobbyists and included a monetary prize pool.“The Gray Swan "Harmful AI Assistant Challenge ", with a prize pool of $40,000, gathered both AI security professionals and hobbyists when it kicked off on January 4th.”Our Interesting Times (substack) ↗
- Seen as collaborator with major safety teamsThird-party writeup states Gray Swan has collaborated with OpenAI and Anthropic on safety-testing and red-teaming, supporting credibility in AI security.“Gray Swan is an AI safety and security company which has collaborated with both OpenAI and Anthropic on safety-testing and red-teaming”Our Interesting Times (substack) ↗
- Fast automated feedback in challengeParticipants reported that automated judging provided immediate feedback, which some preferred to waiting for human review.“the immediate feedback is far superior to waiting for a human review”Our Interesting Times (substack) ↗
- WHOIS / owner identity hiddenScamAdviser flags that the domain owner uses a privacy service and the WHOIS information is hidden, listed as a warning.“The owner of the website is hiding his identity on WHOIS using a paid service”ScamAdviser (app.grayswan.ai) ↗
- Low traffic / low Tranco rankScamAdviser repeatedly notes a low Tranco ranking and that the site 'does not have many visitors', raising a caution for trust at scale.“This website does not have many visitors”ScamAdviser (grayswan.ai) ↗
- Use of iframe / subdomain behaviour flaggedScamAdviser points out the site or its docs/app subdomains include or iframe content from other sites, which the scanner treats as a suspicious signal.“This website is using an iframe or other technology to include content and functionalities located on another website.”ScamAdviser (app.grayswan.ai) ↗
- Registrar associated with many low-score or spammy sitesScamAdviser notes the registrar used by the domain is also used by several sites flagged as low-quality or spammy, which lowered the trust score.“Several spammers and scammers use the same registrar”ScamAdviser (app.grayswan.ai) ↗
- Few or no conventional user reviews / ratingsThere are no reviews or platform ratings on usual third‑party review sites; ScamAdviser pages explicitly show 'No reviews. Be the first to Review.'“No reviews. Be the first to Review.”ScamAdviser (grayswan.ai) ↗
- Automated judging in challenge was finickyParticipants found the automated AI judge finicky early in the challenge, requiring multiple attempts to get submissions accepted (later improved with an appeal system).“All submitted jailbreaks in the arena were automatically reviewed by an AI judge. In practice, this proved quite finicky, as we would often have to spend several messages going back and forth with the model before getting output that would be deemed “harmful enough” to trigger the judge.”Our Interesting Times (substack) ↗
Third‑party signals are mixed: automated site scanners (ScamAdviser) generally label grayswan.ai and its subdomains as probably legitimate and note valid SSL and an algorithmic trust score, but they also flag WHOIS privacy, low traffic, iframe/subdomain behavior, and registrar concerns. Independent writeups around Gray Swan's public red‑teaming challenge portray active community engagement and collaboration with OpenAI/Anthropic, while also reporting friction with the challenge's early automated judging.
- Most third‑party evidence comes from ScamAdviser machine‑scans and a small number of blog/writeup posts; mainstream review platforms (G2, Trustpilot, Capterra) are absent.
- ScamAdviser pages mix human-readable summaries with templated fields (some detail blocks show 'Trust Score 0' in the page template); reported numeric scores above (61, 76) were explicitly stated in the text but represent ScamAdviser algorithm outputs, not user ratings.
- Name collisions exist (unrelated 'iamgrayswan' music site appears in results); only entries clearly tied to the grayswan.ai security company were included.
- No verified customer testimonials, enterprise buyer reviews, or quantitative usage/retention metrics were available in the reviewed texts.
AI Visibility
We ask ChatGPT WITH web search about the company and show its answers plus the sources it cited — the real AI-search experience. The key signal is share of voice: does the company come up when AI is asked about its space, not just when asked by name.
Models asked ChatGPT · web search
Accuracy of what AI said: Accurate — The assistant correctly identifies the founders and leadership, the company's focus on adversarial red‑teaming and runtime protection, and its product names (Shade, Cygnal, Arena).
Gray Swan is a company specializing in AI security, dedicated to enabling the safe and secure use of artificial intelligence. Founded by leading AI safety and security researchers from Carnegie Mellon University, the company focuses on identifying and mitigating vulnerabilities in large language models. (grayswan.ai)
Founders and Leadership:
- Matt Fredrikson: Co-founder and Chief Executive Officer
- Zico Kolter: Co-founder and Chief Scientist
- Rob Jenks: Chief Strategy Officer
- Spencer Whitman: Chief Product Officer
These individuals bring extensive expertise in AI safety and security to the organization. (grayswan.ai)
Services and Products: Gray Swan offers a suite of solutions designed to protect AI systems:
- Shade: An adversarial red-teaming solution that employs LLM-based adversarial agents to test models, guardrails, and deployment contexts, identifying vulnerabilities and providing actionable insights.
- Cygnal: A runtime protection and monitoring solution trained on attacks discovered through Gray Swan's research and global red-teaming network, capable of classifying and blocking adversarial inputs and unsafe outputs in production environments.
- Arena: The world's largest adversarial AI red-teaming network, comprising over 15,000 red teamers who discover novel vulnerabilities daily, providing threat intelligence that informs all of Gray Swan's offerings. (grayswan.ai)
Notable Facts:
- Gray Swan's research has directly informed the safety evaluations of some of the most advanced AI models globally.
- Their work has been cited in published system cards for models such as Claude Opus 4.7, Claude Sonnet 4.6, and GPT-5.
- The company's research has been featured in prominent media outlets, including The New York Times, The Washington Post, TIME, and CNN.
- They have published award-winning papers at top machine learning and AI conferences. (grayswan.ai)
Regarding the company's size and stage, specific details are not publicly disclosed. However, their collaborations with leading AI labs and enterprises, along with their extensive research contributions, suggest a significant presence in the AI security domain.
For more information, you can visit their official website: (grayswan.ai)
The AI agent security landscape features several notable companies and products:
- Okta: Introduced "Okta for AI Agents," a platform to register, monitor, and control AI agents using identity access management systems. (techradar.com)
- WitnessAI: Secured $58 million in funding to specialize in AI security and governance, focusing on safeguarding enterprise AI agents. (axios.com)
- Cato Networks: Acquired Aim Security to integrate AI security solutions into its SASE Cloud Platform, enhancing enterprise AI adoption safety. (itpro.com)
- Certiv: Offers a runtime assurance layer for AI agents, providing full visibility and control over agentic workforces. (certiv.ai)
- Igris Security: Provides a platform for AI-native companies to control their AI stack, including tools for runtime governance and policy enforcement. (igrisecurity.com)
- FortLine Security: Developed AgentShield, offering built-in AI governance and real-time observability for agentic AI environments. (fortline.ai)
- Torq: An autonomous AI SOC platform that automates analyst work across existing security stacks, achieving unicorn status in 2026. (startupstash.com)
- Vectra AI: Utilizes AI for hybrid attack detection, investigation, and response solutions, operating globally since 2011. (en.wikipedia.org)
- OneTrust: Develops governance, risk, and compliance software with a focus on privacy, security, data protection, and AI governance. (en.wikipedia.org)
- Palo Alto Networks: Delivers AI-powered security solutions focused on securing AI models, applications, and autonomous workflows. (marketsandmarkets.com)
- Microsoft: Offers AI-driven security solutions through its cloud and security ecosystem, protecting AI workloads and autonomous systems. (marketsandmarkets.com)
These organizations are at the forefront of securing AI agents, addressing the unique challenges posed by autonomous systems in enterprise environments.
In the rapidly evolving field of AI agent security, several companies have emerged, offering specialized products and services to safeguard AI agents and their interactions. Here are some notable organizations:
Security Solution for AI Agents | Noma Security BotGuard BotGuard provides an all-in-one security platform for AI agents, chatbots, and MCP servers. It offers vulnerability scanning with over 1,000 attack templates, real-time threat blocking through its Shield feature, and a BotGuard Certified badge to demonstrate security compliance. The platform aligns with the OWASP LLM Top 10 standards. (botguard.dev)
NeuralTrust NeuralTrust offers a centralized platform to discover and secure AI agents, enabling safer enterprise adoption. It provides full control over agent behavior and access, from reasoning and planning to action execution. (neuraltrust.ai)
Straiker Straiker focuses on agentic AI security, helping organizations discover AI agents and MCP risks, eliminate threats before deployment, and stop AI agent security incidents at runtime. Their platform includes features like AI agent discovery, adversarial testing, and AI runtime security. (straiker.ai)
Best cybersecurity agentic AI tools for 2025 Rutile Rutile is an AI agent security and governance product that registers every agent as an identity, evaluates tool actions against policy, and creates temporary access, audit evidence, and runtime controls. It is designed for CISO, IAM, security architecture, AI governance, and platform engineering teams. (rutile.tech)
YoroAI - Smart AI Chatbot Platform | Yoroflow™ Oryo Oryo provides an enterprise security control plane for AI agents, offering discovery and security of AI agents, detection of risks across the agentic AI stack, and enforcement of controls to prevent AI agents from becoming security and governance issues. (oryo.io)
These companies are at the forefront of addressing the unique security challenges posed by AI agents, offering solutions that encompass vulnerability scanning, real-time threat mitigation, governance, and compliance.
The assistant provides an accurate, trustable description of grayswan when asked by name, but it did not surface the company unprompted among notable ai‑agent‑security vendors in the category or recommendation queries. For a VC, the named profile is reliable, though grayswan was omitted from the broader vendor lists.
- AI responses can vary between runs and prompts; absence from a surfaced category list may reflect citation choices or index depth, not the company's actual market relevance.
- Public web coverage for early-stage or research‑heavy firms can be limited, so search‑based assistants may omit them from unprompted lists.
Risks & Red Flags
No material public risk signals found. An absence of public concerns isn’t a clean bill of health — early or private companies may simply not generate coverage.
Search of public news, regulator, court, and trade sources returned no material, attributable negative reports about Gray Swan; lack of coverage is common for early/private B2B SaaS companies.
Algorithmic site-trust/checker entries exist for grayswan.ai (e.g., ScamAdviser) but are excluded from risk flags because such services auto-generate scores and are not evidence of misconduct or operational problems.
This assessment covers only publicly-available information; private disputes, undisclosed incidents, or recent developments not yet reported in public sources would not appear here.
Frequently asked questions
What does Gray Swan do?
Enterprise security platform that protects AI agents and models through adversarial evaluation, red‑teaming, and real‑time input/output protection.
How much funding has Gray Swan raised?
Gray Swan has raised $50M in disclosed funding. Its most recent disclosed round was a $40M Series A in 2026, led by Wing Venture Capital and Madrona.
Who are Gray Swan's main competitors?
Gray Swan's named competitors include Enkrypt AI Guardrails, Akto Homegrown AI and GenAI Security, White Circle Control Your AI and Noma Security.
Where is Gray Swan headquartered?
Gray Swan is headquartered in Pittsburgh, Pennsylvania, United States.
What market does Gray Swan operate in?
Gray Swan operates in the Ai agent security market (B2B). Its total addressable market is estimated at USD 13.52 billion (global, projected 2032).
How does Gray Swan's pricing work?
Gray Swan uses a enterprise custom pricing model.
Compiled by AlgoTurk from public web sources · . Not investment advice.