tAI
Cybersecurity Policy
Last updated: August 18, 2026
In short: tAI will not produce working attack code, malware, or unauthorized-access instructions targeting real systems. This check is both part of our general Usage Policy and enforced more strictly on our most capable models (tAI 4.1 and tAI 4.2).
1. Why a separate policy
Cybersecurity topics carry a different kind of risk than general usage violations: output here can be directly "working" code usable against a real system. That's why tAI runs a separate, stricter check on every cybersecurity-related message.
2. Prohibited content
- Working exploit code targeting real systems, networks, or accounts.
- Development of ransomware, viruses, trojans, or similar malware.
- Unauthorized-access tooling, credential theft, phishing infrastructure, or DDoS tools.
- Step-by-step attack instructions aimed at a specific, real target (a person, company, or institution).
- Requests for verifiably working attack code, even when framed as "educational" or an "authorized penetration test," when that authorization can't be verified.
3. What's fine
- General, abstract explanations of security concepts (e.g. "what is SQL injection").
- Defensive security content (logging, monitoring, hardening, security architecture advice).
- Working on your own systems, or a CTF/practice lab clearly under your own control.
- General software security education and code review help.
4. Stricter checks on more capable models
tAI 4.1 and tAI 4.2 are our most capable models and can produce genuinely working, sophisticated code — so the cybersecurity check on these two is noticeably stricter. Other models (tAI 4.0.6P, 4.0.6B, 4.0.7) follow the same underlying rule, but the threshold is somewhat more lenient in borderline cases.
| Model | Cybersecurity check |
|---|---|
| tAI 4.0.6P / 4.0.6B / 4.0.7 | Standard |
| tAI 4.1 / tAI 4.2 | Strict |
That's why a cybersecurity-flagged message shows a "try a different model" option under the warning — in some borderline cases, a less capable model may be able to handle the request.
5. If a violation is detected
Your message is blocked and you'll see: "This message has violated our usage policies and cybersecurity policy." Repeated or clearly malicious attempts may lead to suspension or termination under our Terms of Service.
6. Changes to this policy
We may update this policy from time to time. We'll update the "last updated" date above for meaningful changes.
← Back to tAI