Edge operations

Forward-facing defense, not last year's playbook.

Basalt actively researches and tests the threats most regional providers have not mapped yet — agentic AI, MCP and AI-tool supply chain, post-quantum cryptography, identity-first attack chains, and open-source supply chain compromise. This is the working dossier of what we run against client systems in 2026.

Forward-facing by design. The Basalt practice is structured around the attacker tradecraft that exists right now — not the threat models that comforted boards in 2022.

Agentic AI security

AI agents call tools, fetch data and act on instructions — and the attack surface is more like a software supply chain than a chatbot. We test tool-abuse, indirect prompt injection, agent escape and tool-chain confusion against production deployments.

Questions a Basalt operation will answer

  • Which tools can the agent call and what state can it mutate?
  • How is untrusted input separated from tool instructions?
  • What is the blast radius when the agent is jailbroken — and how is it detected?

MCP and AI-tool supply chain

The Model Context Protocol exposes the same kind of trust assumptions that have burned software supply chains for years — but with less scrutiny. We assess MCP server configs, tool surfaces, sandboxing and version-pinning across your real agent stack.

Questions a Basalt operation will answer

  • Which MCP servers are wired into your agents and where do they run?
  • What does the trust boundary between MCP server and agent actually enforce?
  • How are MCP server updates verified before they hit production agents?

Post-quantum cryptography readiness

Harvest-now-decrypt-later is not a future threat — it is happening now against long-life secrets and signed data. We inventory cryptographic use across your stack, score crypto-agility, and sequence migration to NIST PQC standards (ML-KEM, ML-DSA, SLH-DSA).

Questions a Basalt operation will answer

  • Where does cryptography live across your apps, infra and vendors?
  • How long are your secrets and signed artefacts valuable to an attacker?
  • How will you swap algorithms when (not if) one of them is broken?

Identity-first attack chains

Modern attackers do not break in — they log in. Identity is the new perimeter, and the attack chain runs through federated SaaS, OAuth grants, service principals and CI/CD identities. We model these paths end-to-end and engineer detections for the actual tradecraft.

Questions a Basalt operation will answer

  • Which OAuth apps and service principals have standing access to your tenants?
  • How would you detect a refresh-token theft or session-hijack today?
  • What is the blast radius when one CI/CD identity is compromised?

Open-source supply chain (post-xz, post-tj-actions)

The xz backdoor and tj-actions compromise made the abstract concrete: open-source maintainership is a target. We assess your software supply chain controls, SBOM hygiene, signed-artefact verification and CI/CD trust boundaries against the threats that actually shipped this year.

Questions a Basalt operation will answer

  • What stops a malicious update from a single dependency reaching production?
  • Are your CI/CD secrets scoped, rotated and observed?
  • Do you verify build provenance, or just trust the registry?

Test against current attacker tradecraftScope a frontier-aware engagement

Schedule call