Prove whether a model artifact is safe enough for production intake.
Eresus reviews pickle, PyTorch, ONNX, safetensors, GGUF, archive, and container artifacts from HuggingFace and private model repositories for supply-chain and runtime risk.
Vulnerabilities our researchers found and disclosed
Public advisories credited to the Eresus research team. The same people run your engagement.
This engagement creates value fastest for teams like these.
AI product and platform teams
Teams shipping LLM, RAG, MCP, agent, or model-intake workflows into internal or customer-facing environments.
Security leaders expanding into AI
Organizations that already run pentest programs and now need guardrail, prompt, and tool-abuse validation.
Teams that need explainable hardening
Groups that need policy, prompt, MCP, and runtime findings translated into concrete mitigations and release decisions.
Scope
Risk signals
Outcomes
Evidence standard
Proof you can inspect. Boundaries you can plan around.
Every engagement is bounded by written authorization, evidence handling, and a clear handoff. These are the defaults; the signed scope can tighten them.
Evidence you receive
- Tested prompt, agent, RAG, MCP, and tool-use paths tied to real workflows.
- Redacted transcripts and request/response evidence for validated failures.
- Policy, identity, tool-scope, and rollout remediation guidance.
- Retest status for critical guardrail changes.
Explicit exclusions
- Compliance certification or a claim that the model is safe.
- Training on, retaining, or redistributing customer data or model weights.
- Unapproved destructive tool calls or abuse of third-party systems.
- A guarantee that every prompt or agent path has been discovered.
Review model artifacts as executable supply chain, not passive files.
Problem
We identify where the model came from, how it loads, and where it will run.
Attack scenario
We test malicious pickle, custom operators, archive traversal, secrets, and dependency chains.
Proof
Evidence is tied to file paths, hashes, opcodes, AST nodes, manifests, or model metadata.
Delivery
Intake criteria, quarantine, signing, and retest steps are turned into release decisions.
The questions buyers want answered early.
What AI surfaces do you test?+
Is this just prompt injection testing?+
Do you translate findings into engineering actions?+
We tie risk to business impact.
Findings do not stop at severity labels. We explain which customer workflow, data class, or operational objective is affected.
Deliverables work for engineers and executives.
Engineering teams get reproducible proof and remediation direction; leadership gets the risk narrative, priority, and closure status.
Research and advisories that support this service motion.
Let’s scope this work against the surface that matters most.
Whether this starts as a pilot, a single application, a critical API, an AI agent flow, or a wider program, we start from the highest-impact surface.