AI Governance & Security Vendor Wiki

5 min read

Adaptive Security

Primary category: anti-deepfake.

One-liner — AI-native security awareness training and social-engineering attack simulation (phishing, vishing, smishing, and deepfake voice/video) that drills employees against the kind of AI-generated impersonation attacks they now actually face.

What it does

Adaptive Security is a human-layer defense vendor. It runs simulated social-engineering campaigns against an organization’s own staff — AI-generated phishing emails, SMS (smishing), voice calls (vishing), and deepfake voice/video personas — then measures who falls for them and delivers personalized security-awareness training shaped to each user’s risk profile. It adds real-time threat triage and AI-driven risk scoring. The pitch is that legacy awareness-training tools (KnowBe4-style) were built for generic phishing, whereas attackers can now cheaply clone a CEO’s voice or face, so the training and simulations need to be AI-native too. (Vendor framing — marketing.)

Where it sits in the stack

Sits in anti-deepfake at the foundation layer. This is a people-and-process control, not an inline technical guardrail: it does not sit in the model/prompt path or inspect application traffic. At most it helps screen untrusted input — it hardens the human against malicious inbound social-engineering content — but it is not an inline prompt/egress control at the system level. Trust-zone relevance: the human perimeter (red zone / external untrusted contact), not the AI application data path.

Deployment & architecture

SaaS platform. Delivers simulated attacks across email, SMS, voice, and video channels and reports results; training is delivered to end users. Integrations to verify (IdP/SSO, email/HRIS for user provisioning, SIEM for reporting) — not confirmed from primary sources here. No inline proxy, API-gateway, or model-path component; this is an admin-console + simulation-delivery product, not a runtime security control.

Positioning & differentiators

Known for being the OpenAI-backed, AI-native entrant in security awareness training, explicitly targeting deepfake and multi-channel (voice/video/SMS) social engineering rather than email-only phishing. Differs from digital-risk-protection peers like doppel (which scans the external internet for impersonation of your brand/executives) — Adaptive points inward at your own employees, training them. Differs from media-authentication vendors reality-defender, getreal, and pindrop, which detect whether a given piece of media or a call is synthetic; Adaptive instead uses synthetic media offensively to test and train staff. Named customers (vendor-stated): PayPal, Xerox, Bose, NHL, PGA, Figma, Ramp, Vimeo, Perplexity.

Ownership, funding & M&A

Independent, VC-backed. No acquisition (seed carried no M&A flag; confirmed none found). Founded 2025 (public launch January 2025); HQ New York; co-founders Brian Long (CEO) and Andrew Jones.

Funding trajectory, all confirmed against primary releases:

  • Series A — $43M, announced 2025-04-02, co-led by Andreessen Horowitz (a16z) and the OpenAI Startup Fund (noted as OpenAI’s first cybersecurity investment).
  • Series A extension — to $55M total, announced 2025-09-09, via an OpenAI Startup Fund follow-on.
  • Series B — $81M, announced 2025-12-16, led by Bain Capital Ventures, with NVentures (NVIDIA), OpenAI Startup Fund, a16z, Abstract Ventures, Capital One Ventures, Citi Ventures. Total raised $146.5M.

The brief’s special-check item — “verify the OpenAI-led funding claim” — is confirmed: the OpenAI Startup Fund co-led the April 2025 Series A and extended it in September 2025 (primary PR Newswire releases). Note nuance: OpenAI is an investor, not the sole “lead” of the original round (a16z co-led) and not an owner.

CTO / hedge-fund lens

Day-2, medium fit. Deepfake-enabled vishing/BEC against finance staff (fraudulent wire approvals, fake-CEO calls) is a real and rising threat for asset managers, so AI-native awareness training and deepfake-call simulation are reasonable for a fund with meaningful headcount and treasury/payment operations. But this is a human-risk control, not part of the AI-application security stack the rest of this wiki covers, and it competes with incumbent awareness-training the fund may already license. No direct SR 11-7 / model-risk relevance. Small funds (sub-50 people) may get more leverage from payment-process controls (callback verification, dual approval) than from a dedicated training platform. Young company (founded 2025) — diligence product maturity and contract terms.

Competitors / alternatives

  • doppel — external digital-risk protection / impersonation detection (outward-facing vs Adaptive’s inward training).
  • reality-defender, getreal, pindrop — deepfake/synthetic-media detection (detection vs training).
  • Incumbent security-awareness-training vendors (e.g. KnowBe4, Proofpoint) — not yet covered here; the legacy category Adaptive is attacking.

Open questions / to verify

  • Exact integration set (IdP/SSO, SIEM, HRIS) — not confirmed from primary sources.
  • Pricing / packaging and whether deepfake-call simulation is a separate module.
  • Real-world efficacy vs incumbent awareness-training (independent evidence, not vendor claims).

Sources

History

  • [2026-06-28] Stub created from seed registry.
  • [2026-06-28] Researched; confirmed OpenAI Startup Fund co-led/extended Series A (primary PR Newswire, Apr & Sep 2025) and Bain-led $81M Series B (Dec 2025, $146.5M total); independent (no M&A); founded 2025, HQ New York, founders Brian Long & Andrew Jones; SaaS human-layer training/simulation; set hedge_fund_fit medium, risk role untrusted-input screening, ownership_confidence high.

Graph View

Backlinks