Riptides

We were speculatively building kernel drivers for 540+ variants every night. Not every kernel shows up on day one, but when it does, the node can't wait for the next nightly run. So we flipped the model: build on first encounter, cache forever, and let the fleet shape the build matrix automatically. If you're distributing kernel modules, eBPF programs, or any out-of-tree driver across a diverse fleet the patterns here (Koji fallback chains, livepatch discovery, distro-version-pinned builds, demand-driven CI) might save you a few painful debugging sessions. Read on 👇 https://lnkd.in/dEqFnhMa

We just open-sourced KeyLedger, a tool that gives you a unified inventory of every AI API key issued across your AI providers. Health scoring, snapshot diffs, encrypted credential storage, and an interactive TUI — no database, no infrastructure, one binary.  https://lnkd.in/g-Kt8P2s

Anthropic’s new workload identity federation support for Claude fits very naturally into Riptides. Workloads can securely communicate with Claude without managing long-lived Anthropic API keys directly, while Riptides transparently handles federation, short-lived credential acquisition, rotation, revocation, and runtime enforcement underneath. Runtime identity in. Short-lived credentials out. https://lnkd.in/duaXsRUy

Modern infrastructure still makes a surprisingly dangerous assumption, if the token is valid, the action is probably legitimate. Our latest post explores why token validity is the wrong guarantee for workloads access, and why runtime legitimacy matters far more than most infrastructure currently validates. https://lnkd.in/dssbx8Ui #Security #RuntimeSecurity #ZeroTrust #SPIFFE #WorkloadIdentity

Someone may already be recording your #mTLS traffic. Not to read it today, but to decrypt it when quantum Q-day arrives. We broke down why workload identity is the high-value target, where the ecosystem stands, and what we're doing about it at Riptides. https://lnkd.in/dzDjjWfm #PostQuantumCryptography #PQC #QuantumSecurity #QuantumComputing #TLS #mTLS

#SPIFFE gives AI agents the cryptographic, ephemeral identity they need but #SPIRE was never designed to deliver it at the agent layer. We break down why user-space identity issuance, sidecar architectures, and manual certificate lifecycle fall apart for polyglot, dynamically spawning agents and how kernel-level SVID injection, secretless credential management, and per-process policy enforcement close the gap. https://lnkd.in/ddDNwE8S #SPIFFE #WorkloadIdentity #AgenticAI #MCP #ZeroTrust #NonHumanIdentity

TLS 1.2 doesn't mandate forward secrecy - every recorded internal session is a liability if a certificate is later compromised. We upgraded the Riptides kernel module to enforce TLS 1.3 on all internal mTLS, with opt-in hybrid post-quantum key exchange (X25519 + ML-KEM-768). One config change, every workload, no application code touched. https://lnkd.in/dgtCgYD6

Remote MCP servers put OAuth access tokens in agent memory, plaintext, replayable, one prompt injection away from exfiltration. Our fix: an intermediate authorization server that brokers the real flow and injects credentials at the kernel level. The agent never possesses the token that grants access. https://lnkd.in/dQYbCWSX

AI agents are transforming productivity, but are we granting them more access than they truly need? Our latest blog post, "Our AI is helpful. Also, slightly overprivileged," explores the critical balance between AI utility and system security. https://lnkd.in/dPgdV5M4

Identity over Secrets: Stop managing credentials. Neither your AI agents nor your developers should be handling raw credentials. They should be granted ephemeral access, not static keys. In this demo, Claude connects to Jira via MCP using centrally managed policies. Access is dynamically evaluated by combining the Human Identity with the Agent Identity. Manage access, not secrets. Interested? Ping us.