The prompt ban did not see the path
Gambit and Forbes documented AI harnesses walking into online retailers for about $25 a target. Banning a model account does not map the authority path those agents took.
Resources for your next security decision.
Choose a starting point, understand how the work runs, and know what your team will receive. Start with the question you need answered.
Find the answer you need.
Each guide answers a different question about working with Serpio. All are open to read.
Where should we start?
Compare adversarial testing, secure design review and the other services. Find the starting point that fits your system and the problem you need to solve.
How will the work run?
Walk through scoping, testing, findings and retesting. See how coverage and severity are measured, and what must be agreed before anything is touched.
Which risks can you test?
Read the attack classes we test against, from prompt injection and unsafe tool use to identity, supply-chain and binary risks. Use the register to discuss coverage for your system.
What will my team receive?
See how Serpio Atlas maps the system in scope and connects findings to the paths that made them possible. The map is yours to keep after the engagement.
What can I show an auditor?
Understand the findings, coverage record and framework mapping in the evidence package. See what that evidence supports and where its limits lie.
Before the first conversation.
A useful starting point is a system diagram, the concern that prompted the review, and any constraints on testing.
- Your system
- The applications, APIs, agents or infrastructure you want reviewed, and how they connect.
- Your concern
- A proposed change, a known weakness, or something your team needs to prove.
- Your constraints
- Environments, test windows, data handling and any air-gap requirements.
- Your decision
- What you need to decide or change after the work is complete.
How we think about the risks.
Our published notes on agent permissions, testing and the systems around them.
Authority paths, not prompt filters
Prompt firewalls watch text and cloud posture tools watch infrastructure. Serpio Atlas records what each agent may call and reach.
Report a vulnerability.
Found a defect in a Serpio system? Send the affected URL or system, steps to reproduce, and the impact you observed to security@serpio.ai.
We acknowledge reports and credit you unless you prefer to remain anonymous. We do not threaten good-faith researchers. An acknowledgment confirms receipt; it does not promise a fix or a date.
Our machine-readable reporting details are in security.txt. For questions about your own systems or an engagement, contact the team.
Have a system in mind?
Tell us what you run and what you want to understand. A senior operator replies at hello@serpio.ai.