Enkrypt AI

Primary category: ai-runtime-security. Also ai-red-teaming.

One-liner — A full-lifecycle LLM/agent security startup that pairs automated red-teaming (find the holes) with runtime guardrails (an AI firewall that blocks them), wrapped in compliance/risk reporting mapped to frameworks like the EU AI Act and NIST.

What it does

Enkrypt AI sells the “test-then-protect” loop for generative AI. On the red-teaming side it runs automated adversarial attacks against models, apps, and agents to surface prompt injection, jailbreaks, data leakage, hallucination, bias, and (in its published research) CBRN and agent-misuse risks. On the runtime side, its guardrails / “Sentry” control layer sits between users and the model and inspects and blocks unsafe inputs and outputs and policy violations in real time. A distinguishing pitch is policy-to-rule automation: it converts human-readable compliance policies into enforced guardrail rules, plus continuous risk reporting and audit evidence. The company also publishes an “LLM Safety Leaderboard” benchmarking 200+ models, which doubles as marketing and as model-selection input.

Recent product expansion is heavily agentic: Agent Red Teaming, Agent Guardrails, an Agent Policy Engine, an MCP Scanner (scan MCP servers for risk), and an MCP Gateway. It was named a Gartner Cool Vendor in AI Security (2025, per the vendor).

Where it sits in the stack

Primary home is ai-runtime-security (the AI firewall / runtime guardrail layer, model-prompt layer, Day-1 if you put GenAI in front of users or data). Its testing side belongs to ai-red-teaming, and its MCP/agent tooling reaches into mcp-gateway and agent security territory.

Lethal-trifecta role: the runtime guardrail is one of the few controls that touches all three legs — it screens untrusted input (prompt-injection/jailbreak detection), watches for sensitive-data exposure in prompts/outputs (PII/leakage filtering), and can block unsafe egress (tool calls / outbound content). Trust zones: it lives at the boundary between the untrusted user/web zone and the model.

Deployment & architecture

SaaS platform consumed via API and SDK, with the guardrail deployable as an inline control layer / proxy between application and model so it can inspect traffic in real time. Integration surface emphasizes MCP (scanner + gateway for agent tool access) and multi-model coverage (guardrails sit in front of many model providers). Compliance reporting maps detections to EU AI Act / NIST. (SIEM/IdP/DSPM integration depth not confirmed from primary sources — see open questions.)

Positioning & differentiators

Enkrypt’s angle is breadth across the lifecycle — red-teaming, runtime guardrails, and compliance reporting in one place — versus point players. Compared to neighbors:

• vs lakera and prompt-security — Enkrypt leans harder into the red-teaming + compliance-reporting story, not just inline guardrails.
• vs splxai, mindgard, promptfoo — those are more red-teaming/eval-first; Enkrypt also ships the production firewall.
• vs pillar-security, hiddenlayer, witnessai, prisma-airs, calypsoai — these overlap on runtime/AI-firewall; Enkrypt is a smaller, independent, research-forward startup rather than a platform or acquirer.
• Known for visible research output (CBRN red-teaming study, guardrail-bypass reports on Azure/AWS/Meta, agent risk taxonomy) and the LLM Safety Leaderboard.

Ownership, funding & M&A

Independent, VC-backed. Confirmed $2.35M seed (Feb 2024), led by Boldcap, with Berkeley SkyDeck, Kubera VC, Arka VC, Veredas Partners, and Builders Fund. No M&A — not acquired, has not acquired anyone (no seed flag; nothing found).

The research brief hinted at a ”~$8M Series A (2025, led by Boldcap)” and “MIT origins.” Neither was confirmable against primary sources as of 2026-06-28: the only publicly disclosed round is the seed, and the founders are Yale PhDs (Sahil Agarwal, CEO; Prashanth Harshangi, CTO), not MIT. Some third-party aggregators cite a larger Series A figure but without a primary announcement. Treated as unconfirmed, not asserted. (soft mismatch, non-blocking.)

CTO / hedge-fund lens

The runtime-guardrail function is Day-1 for any fund putting an LLM or agent in front of users or sensitive data; the red-teaming and compliance-reporting functions are Day-2 maturity but become Day-1 under a model-risk regime. For an asset manager, the compliance-mapping and audit-evidence angle (EU AI Act / NIST, and by extension SR 11-7-style model-risk documentation) is the most differentiated reason to look here — it turns guardrail enforcement into auditable evidence. Caveat: Enkrypt is an early-stage, small independent vendor, so vendor-risk/longevity diligence matters more than with a Palo Alto (prisma-airs) or Cisco. Fit: medium — attractive capability set and compliance framing, tempered by startup maturity and the fact that larger platforms now bundle similar guardrails.

Competitors / alternatives

lakera, splxai, pillar-security, hiddenlayer, prisma-airs, witnessai, mindgard, promptfoo, calypsoai, prompt-security.

Open questions / to verify

• Series A: amount, date, lead — could not confirm a 2025 Series A; only the $2.35M seed is documented. Re-check for a later announcement.
• “MIT origins” appears incorrect (founders are Yale PhDs) — confirm/retire this alias note.
• Depth of enterprise integrations (SIEM/SOC, IdP, DSPM) beyond MCP and model providers.
• Whether guardrails run self-hosted/on-prem for regulated buyers, or SaaS-only.

Sources

• Enkrypt AI homepage — fetched 2026-06-28 — supports: product portfolio (Agent Guardrails/Red Teaming/Policy Engine, MCP Scanner/Gateway), SaaS+API+SDK deployment, EU AI Act/NIST mapping, Gartner Cool Vendor 2025; confidence: med (vendor marketing).
• Enkrypt AI vendor profile — IT-Harvest — fetched 2026-06-28 — supports: founded 2022, Boston HQ, Yale-PhD founders, $2.35M seed (Boldcap), Sentry inline control-layer description; confidence: med.
• Enkrypt AI secures $2.35M — Tech Startups — fetched 2026-06-28 — supports: seed amount/date/investors, founders, founding year, HQ, Sentry product; confidence: high (for the seed round).

History

• [2026-06-28] Stub created from seed registry.
• [2026-06-28] Researched; established founded 2022, Boston HQ, Yale-PhD founders (Sahil Agarwal/Prashanth Harshangi), independent with confirmed $2.35M seed (Feb 2024, led by Boldcap). Could NOT confirm the brief’s ~$8M 2025 Series A or “MIT origins” — flagged as unconfirmed (soft). Set deployment saas/api/sdk/inline-proxy, trifecta all three legs, hedge_fund_fit medium. Cached 3 sources. status: researched.