Skip to content

Market, foundations and acquisitions (as of 2026-11-18)

Star counts are from the GitHub API on 2026-11-18 unless marked otherwise.

The spec side of agent identity moved fast in 2026: WIMSE AIMS was adopted, Transaction Tokens is near publication, and ID-JAG is stable in the MCP Enterprise-Managed Authorization extension. The open-source relying side is thin. Almost everything ships issuance or a vendor gateway, not neutral verification of the whole call chain.

CNCF has no project for agent or workload identity beyond SPIFFE/SPIRE, while its own AI TCG and TOC initiatives say interoperable agent identity is “currently lacking”.

Identity and agent-security acquisitions in 2026 shifted toward agent/NHI identity and MCP gateways. OSS-led targets sold cheaply relative to closed NHI vendors: stars got them acquired but did not raise the price.

Area State Gap
WIMSE issuance SPIRE v1.16.0 (2026-11-05) issues WIT-SVIDs (experimental); go-spiffe has exp/svid/witsvid None on issuance
WIMSE verification (WPT, HTTP-sig, mTLS) wimsey (Rust, Apache-2.0) and one stalled PoC; go-spiffe WPT PR #415 still a draft Nothing deployable at an enforcement point
Transaction Tokens tokenetes (CNCF Sandbox, 66 stars, maintenance mode, K8s and SPIFFE only), aramase/kontxt (26 stars, single maintainer) No neutral, actively maintained Txn-Token Service plus verifier
ID-JAG issuance Dex (merged 2026-11-11, unreleased), Casdoor v4.27.1+, commercial IdPs Receiving side and the agent hop after it
AIMS reference implementation None; the draft has no Implementation Status section Entire composition
Delegation-chain verification Per-draft author libraries (attenu-guard, Tenuo, mission-core) Format-neutral verifier wired to Txn-Tokens and a PDP

Already well covered and not worth competing on: SD-JWT VC and status lists, DPoP, CIMD, AuthZEN PDPs, MCP auth proxies.

  • Identity and security projects accepted 2024-2026: Kuadrant (2024-07), Tratteria/Tokenetes (2025-01), kagent (2025-05), Cedar (2025-10), OAuth2-Proxy (2025-10), OpenShell and Agent Substrate (2026-10). OpenFGA moved to Incubating (2025-11); cert-manager (2024-11) and Kyverno (2026-03) graduated.
  • agentgateway went to the Linux Foundation (2025-08) and then the Agentic AI Foundation (AAIF). OpenBao moved to OpenSSF (2025-06).
  • Sandbox process: issue form, TOC review roughly every two months, 66% gitvote. Requirements include Apache-2.0 and a MAINTAINERS file with an organization column. Median time to approval since 2025 is about 73 days. Common rejections: insufficient momentum (“grow the community for 6-12 months”), missing files, single-vendor runtime lock-in, overlap with existing projects.
  • Stated demand: the CNCF AI TCG “Cloud native agentic standards” (2026-03) asks for a unique workload identity per agent instance, short-lived credentials and re-verification per operation; TOC initiatives on MCP server authN/authZ (toc#1890) and agentic foundations (toc#1746) are open. AAIF leadership named identity, attestation and audit as open problems at AGNTCon 2026.
  • Likely home: CNCF Sandbox next to SPIFFE/SPIRE and Tokenetes, with specification work in IETF WIMSE / OAuth and the OpenID Foundation.
Acquirer and target Date Price Notable OSS
Palo Alto Networks and CyberArk closed 2026-02 about $25B none
CrowdStrike and SGNL (Transaction Tokens authors) announced 2026-01 about $740M tokenetes lineage
ServiceNow and Veza closed 2026-03 about $1.2B none
Cisco and Astrix announced 2026-05 reported $250-400M none of note
Snowflake and Natoma (MCP gateway) announced 2026-05 reported $110-128M none
Palo Alto Networks and Portkey closed 2026-05 $117M Portkey gateway, about 12k stars at the deal (secondary source)
Vercel and Better Auth 2026-07 undisclosed better-auth, 28k+ stars
Okta and Permiso closed 2026-08 reported under $200M none
Cyera and Oasis (NHI) closed 2026-09 about $1B none
Snyk and Invariant Labs 2025-06 undisclosed mcp-scan (now snyk/agent-scan)
OpenAI and Promptfoo announced 2026-03 undisclosed promptfoo

What the deals show:

  • Most targets in this space were closed-source NHI, PAM or ITDR vendors bought for revenue.
  • OSS-led targets were a dominant developer library, a developer-loved gateway or CLI on AI traffic, or a research team whose scanner backed its disclosures.
  • Price tracked revenue, not stars.
  • Better Auth was the only auth library to reach large scale (30.8k stars), by filling a developer-experience gap as Auth.js stalled.
  • MCP security tools spread through security research (Invariant’s tool-poisoning disclosures behind mcp-scan; Koi’s postmark-mcp report) rather than launches.
  • Foundation-backed gateways (agentgateway, IBM ContextForge) grew through vendor endorsement with almost no Hacker News traction.
  • Stand-alone MCP auth helper libraries stayed small because the official SDKs absorbed them. New work here has more room as a gateway or broker than as a library.