Lakera

Researched 2026-06-28; re-verified 2026-07-27. Primary category: ai-runtime-security. Acquired by Check Point Software (NASDAQ: CHKP) — closed 2025-10-22 for ~$201.8M. The Lakera brand is still live (lakera.ai loads unchanged, with a Check Point copyright footer), so it remains buyable under its own name — but as a Check Point product line, not an independent vendor.

One-liner — The AI-native runtime-security pure-play of “Gandalf” fame (real-time prompt-injection / GenAI defense), now Check Point’s AI-security engine and the foundation of its Global Center of Excellence for AI Security.

What it does

Lakera protects LLMs, GenAI apps and autonomous/agentic systems at runtime. Its core product, Lakera Guard, screens inputs and outputs in real time against prompt injection, jailbreaks, data leakage, and model manipulation across multiple AI frameworks. Its threat intelligence is fed partly by Gandalf, a viral prompt-injection education game (>1M players; >50M data points) used even by Microsoft for security training. Lakera also does AI red teaming.

Where it sits in the stack

Primary ai-runtime-security (inline AI firewall / guardrails) with a secondary ai-red-teaming tag. Layer: model-prompt. Lethal-trifecta role: strongest on untrusted input (it is the best-known prompt-injection defender) and sensitive data (output/data-leak screening). Sits at the green↔red boundary in front of model endpoints.

Deployment & architecture

SaaS and self-hosted; primarily API-based (low-latency guardrail calls around model requests) plus SDK integration — model- and cloud-agnostic. Post-close, Lakera combines with Check Point Infinity / GenAI Protect and forms the foundation of Check Point’s Global Center of Excellence for AI Security. Integrations: AI gateways, the app’s LLM pipeline, SIEM.

Positioning & differentiators

Lakera is the category’s best-known prompt-injection specialist, with a research-and-community moat (Gandalf, an open prompt-injection dataset) and a lightweight API-first guardrail rather than a heavy inline proxy. Founders came from Google/Meta AI.

Since 2025-10-22, read that positioning as a platform vendor’s, not a pure-play’s. Lakera is no longer an independent, model-neutral third party you can buy without a relationship to a network-security incumbent; it is how Check Point competes in AI security. That moves it out of the “best-of-breed specialist” bucket and into the same bucket as prisma-airs (Palo Alto), cisco-ai-defense (Cisco), F5’s calypsoai, Cato’s aim-security and SentinelOne’s prompt-security — the 2025 wave in which nearly every major network/security platform bought its AI guardrail rather than building one. What survives the deal is the technical differentiation (the prompt-injection research and Gandalf-fed threat intel are genuinely ahead of most of that peer group); what does not survive is the neutrality argument.

Alternatives still independent as of this page’s last check, if vendor-neutrality is a requirement: witnessai, enkrypt-ai, mindgard, hiddenlayer. The pure-play population in this category is thinning fast; assume any name on that list can be acquired inside a buying cycle.

Ownership, funding & M&A

Founded 2021 by David Haber (CEO), Dr. Mateo Rojas-Carulla and Dr. Matthias Kraft; dual HQ Zurich + San Francisco. Funding ~$30M total, headlined by a $20M Series A (July 2024) led by Atomico (Citi Ventures, Dropbox, redalpine).

Acquired by Check Point Software (NASDAQ: CHKP). Announced 2025-09-16; closed 2025-10-22 — Check Point “completed the acquisition of all outstanding shares of Lakera AI AG,” per Note 3 of its audited FY2025 Form 20-F. Sign-to-close was 36 days. Confidence: high (issuer’s own audited SEC filing, not press).

Price: ~$201.8M total consideration, also from the 20-F. This is materially below the ~$300M that press (Calcalist and downstream coverage) reported at announcement and that this page previously carried. The two are not necessarily in conflict — accounting “total consideration” excludes amounts that must be expensed as post-close compensation, and retention/earn-out packages are routinely folded into the number journalists are briefed on — but no primary source itemizes the ~$100M gap, so treat $201.8M as the authoritative purchase price and ~$300M as an unofficial figure of uncertain composition.

Contradiction (soft / scope-mismatch): audited total consideration $201.8M (Check Point 20-F, FY2025) vs. ~$300M reported at announcement (Calcalist). Different measures, most likely purchase price vs. purchase price plus retention. Noted 2026-07-27; using the filed figure.

The purchase price allocation is worth reading, because it telegraphs intent: goodwill is $150.1M — 74% of the price — “primarily attributed to synergies,” i.e. this was bought for technology and team fit, not for Lakera’s revenue. Core technology got a 7-year useful life, but the Lakera trademark and its customer relationships were each assigned a 1-year useful life. Check Point’s own accounting assumption is that the brand and the standalone customer base have about a year of remaining value from October 2025, while the engine has seven — the standard fingerprint of a plan to absorb the product and retire the standalone identity. (The useful lives are filing facts; the read on intent is ours.)

Brand status (checked 2026-07-27)

lakera.ai returns 200 with no redirect to checkpoint.com, and the page title is still “Lakera: The AI-Native Security Platform to Accelerate GenAI” — independent-vendor marketing, no Check Point co-branding above the fold. The tell is the footer: “©1994-2026 Check Point Software Technologies Ltd.” — 1994 is Check Point’s founding year, three decades before Lakera existed, so that copyright line is the parent’s boilerplate on a property it owns. So: ownership changed 2025-10-22, the marketing site has not. The wiki keeps lakera as the slug and lakera.ai as the website, and the survey keeps “Lakera (Check Point)” as the option. Given the 1-year trademark life in the 20-F, expect this to need re-checking inside a year.

CTO / hedge-fund lens

Day-1 control type — a runtime AI firewall focused on prompt injection is a baseline control for any shop exposing LLMs to untrusted input. Lakera’s API-first model suits teams that want guardrails in code without standing up an inline proxy. Indirect SR 11-7 relevance (control + red-team evidence).

Post-close, the practical questions are commercial rather than technical:

• You are now buying from Check Point. Procurement, contracting, support escalation and roadmap influence run through a large public platform vendor (NASDAQ: CHKP, ~$678M of revenue in Q3 2025 alone). If you are already a Check Point shop, that is a simplification and you should ask about bundling into your Infinity agreement. If you are not, you are opening a vendor relationship with a network-security incumbent to get a guardrail API — weigh that against a still-independent alternative.
• Standalone availability is not yet in doubt, but is not guaranteed either. The product is still sold under its own brand and site as of 2026-07-27. But Check Point amortizes the trademark and the acquired customer relationships over one year, which is what a company does when it expects the standalone line to be substantially absorbed. Contract accordingly: ask for multi-year pricing, renewal protection, and written commitment on continued availability of the standalone API rather than assuming today’s packaging persists.
• The technology risk is low and the counterparty risk went down. Unlike an acqui-hire, this was a $201.8M cash purchase of a product line with a 7-year technology life and a Center-of-Excellence mandate — Check Point intends to invest in the engine. For a fund, the acquisition arguably reduces the “will this startup exist in three years” concern that applies to the remaining pure-plays.

Still a good fit for funds building their own LLM apps/agents; just diligence it as a Check Point product with a legacy brand, not as a Swiss startup.

Competitors / alternatives

prompt-security, calypsoai, witnessai, aim-security, prisma-airs, cisco-ai-defense, enkrypt-ai, mindgard, hiddenlayer.

Open questions / to verify

• Confirm the deal closed — resolved 2026-07-27: closed 2025-10-22 per Check Point’s FY2025 20-F.
• Official deal value — resolved 2026-07-27: ~$201.8M total consideration per the 20-F. Still open: what the ~$100M gap to the reported ~$300M consists of (presumed retention/earn-out, unconfirmed).
• How long the Lakera brand survives. Site is unchanged as of 2026-07-27, but the 20-F assigns the trademark a 1-year useful life. Re-check for a rebrand to a Check Point product name.
• Whether Lakera Guard remains purchasable standalone, outside a Check Point Infinity agreement, and at what packaging/pricing.
• Whether the founders (David Haber et al.) are still at Check Point running the Global Center of Excellence for AI Security.

Sources

• Check Point Software Technologies Ltd., Form 20-F for FY2025 (SEC), Note 3: Acquisitions (f) — filed 2026-03-31, fetched 2026-07-27 — supports: close date 2025-10-22, “completed the acquisition of all outstanding shares of Lakera AI AG”, total consideration ~$201.8M, purchase price allocation (goodwill $150.1M; core technology $44.0M / 7 yrs; customer relationships $4.4M / 1 yr; trademark $0.3M / 1 yr); confidence: high (audited issuer filing)
• Check Point Q3 2025 results, Form 6-K (SEC) — filed 2025-10-28, fetched 2026-07-27 — supports: CEO Nadav Zafrir framing Lakera as advancing the “AI First strategy”; Q3 2025 revenue $678M; confidence: high
• lakera.ai homepage — fetched 2026-07-27 — supports: brand still live (HTTP 200, no redirect, original title) with a “©1994-2026 Check Point Software Technologies Ltd.” footer; confidence: high
• Check Point Acquires Lakera… (Check Point press release) — fetched 2026-06-28 — supports: 2025-09-16 announce, founders, founding 2021, Zurich/SF HQ, Center of Excellence integration, then-expected Q4 2025 close; confidence: high
• Lakera raises $20M Series A (TechCrunch) — fetched 2026-06-28 — supports: $20M Series A, ~$30M total, Atomico lead, Gandalf; confidence: high
• Check Point acquires Lakera in $300 million deal (Calcalist) — fetched 2026-06-28 — supports: the ~$300M figure reported at announcement; confidence: low-med (superseded on price by the 20-F’s $201.8M; retained to document the discrepancy)
• Cached: raw/sources/2026-07-27--lakera--checkpoint-20f-close-date-and-price.md

History

• [2026-06-28] Stub created from seed registry.
• [2026-06-28] Researched; confirmed acquired by Check Point Software (announced 2025-09-16, ~$300M reported, expected close Q4 2025). Seed flag verified. Filled founding (2021), Zurich/SF HQ, founders, ~$30M funding, Gandalf/prompt-injection positioning. Confidence high (close pending per release).
• [2026-07-27] Deal closed — acquired-pending → acquired-closed. URL audit found lakera.ai live and unchanged under the Lakera brand but carrying a “©1994-2026 Check Point Software Technologies Ltd.” footer. Went to the acquirer’s SEC filings rather than press: Check Point’s audited FY2025 Form 20-F (Note 3(f)) states it “completed the acquisition of all outstanding shares of Lakera AI AG” on 2025-10-22 for total consideration of approximately $201.8M — correcting both the null close date and the ~$300M press price this page had carried. Recorded the price discrepancy as a soft/scope-mismatch contradiction (audited consideration vs. a press figure that likely includes retention). Added the purchase price allocation and the signal in it: 74% goodwill, 7-year technology life but 1-year trademark and customer-relationship lives. Kept website: https://www.lakera.ai and the lakera slug — brand persists, no rename. Rewrote positioning (Lakera is now a platform vendor’s product, grouped with prisma-airs / cisco-ai-defense / calypsoai / aim-security / prompt-security, not with the remaining pure-plays) and the CTO lens (buy-from-Check-Point implications, contracting advice, standalone-availability risk). verify_after cleared; new open questions on brand survival and standalone packaging.