AI Use & Responsible AI Policy

Jettson.dev gives developers a runtime for autonomous AI agents. This policy explains what those agents are, what they can do, what they cannot do reliably, and what you must do when using them.

1. Models in use

Jettson agents reason using one or more large language models, including:

• Anthropic Claude (the current default for autonomous reasoning).
• OpenAI GPT models (used selectively for specific workflows).
• Jettson 6.0 — our internal product name for the agent reasoning configuration; the underlying model providers, versions, and parameters are pinned server-side and may change without breaking your code.

The specific provider and version routing is operational and may change. We will not silently degrade quality for cost reasons.

2. What agents can do

A Jettson agent runs inside its own Linux environment and may:

• Read its task prompt and the user memory recalled at start.
• Browse the public web (headless browser).
• Make outbound HTTP requests.
• Read and write files in its workspace.
• Run shell commands.
• Install software needed for the task.
• Write to long-term memory that persists across runs.

3. Autonomous action — your responsibility

Agents take real actions on real systems. Once you start an agent with a goal, the agent decides which tools to call, what to type, what to install, what to post, and how to interpret the output it receives. These actions can have real consequences in the world (sent emails, posted comments, made API calls, written files, triggered webhooks, spent money on a connected account, etc.).

You are responsible for: the tasks you assign; the credentials you provide; the third-party integrations the agent reaches; the supervision of the agent's output; and any harm caused by its actions. Jettson is not liable for agent decisions, errors, hallucinations, or unintended actions.

4. Hallucinations and accuracy

LLMs produce confident-sounding output that may be inaccurate, fabricated, outdated, or biased. Do not rely on agent output as fact without independent verification. Treat outputs as drafts to be reviewed, not as decisions to be executed.

5. Not professional advice

Agent output is not legal, medical, financial, tax, or other professional advice. Do not use Jettson to deliver such advice to end users without a qualified professional in the loop.

6. Bias and fairness

The underlying models reflect biases present in their training data. Outputs about people, demographics, jobs, or controversial topics may be unfair, stereotyped, or wrong. You must review and adjust outputs before relying on them, particularly when an agent's output affects an individual's access to opportunities, services, or rights.

7. Human oversight requirement

For any agent task that produces consequential decisions about individuals — including but not limited to credit, employment, housing, insurance, education, healthcare, immigration, or judicial outcomes — a qualified human must review the output before it is acted upon. Using Jettson to fully automate such decisions is prohibited by the Acceptable Use Policy.

8. Prohibited high-risk uses

See the Acceptable Use Policy for a complete list. Highlights:

• No CSAM or content sexualizing minors.
• No non-consensual intimate imagery.
• No content promoting violence, terrorism, genocide, or self-harm.
• No election misinformation or political impersonation.
• No fraud, market manipulation, or surveillance.
• No covered HIPAA data without a BAA (none currently available).

9. Training on your data

We do not train base models on your prompts, agent outputs, or memory content without your explicit opt-in. We may use aggregated, de-identified usage telemetry (counts, latencies, error rates) to operate and improve the Service.

Our upstream model providers (Anthropic, OpenAI) have their own data policies. Anthropic does not train on API data by default. OpenAI does not train on API data by default. We rely on these contractual commitments; if a provider's posture changes we will update this page and notify Accounts.

10. Kill switch

You can stop a running agent at any time:

• In the dashboard: open the agent and click Stop.
• Via API: DELETE /api/v1/agents/{id}.
• Via SDK: await jettson.agents.cancel(agentId).

Stopped agents transition to stopped immediately; cleanup of the container completes within ~5 seconds.

11. Watermarking and disclosure

We do not currently watermark agent outputs. If your downstream context requires disclosure that content was generated by an AI agent, you are responsible for that disclosure.

12. Reporting AI harm

If you encounter agent behavior that is unsafe, illegal, or causes harm, report it to customertek@rwxtek.com with subject "AI safety report" and include the agent ID, your account email, and a description of what happened. We investigate every report.

13. Contact

AI policy questions: customertek@rwxtek.com.