Jailbreak (in AI context)
Related service Cybersecurity
DEFINITION
Bypassing an LLM's safety guardrails via role-play, multi-step manipulation, or low-resource languages. Public lists now carry 2000+ known patterns.
- Threat model→
A structured exercise that walks the system's actors, attack surface, risks, and controls. Day one of every DField project · before any code.
- Penetration test (pentest)→
Manual + tooled attack simulation that reveals what an attacker could achieve. We deliver findings as PRs in your repo, not an 80-page PDF.
- DevSecOps→
Security as a continuously-running CI step (SAST, DAST, SCA, IaC scan), not an annual project. Runs against every push; every sprint closes at least one security bug.
- MFA (Multi-factor auth)→
Two or more factors (TOTP, WebAuthn, biometric) beyond a password. Table-stakes in SaaS today · enterprise procurement disqualifies you without it.
- SOC 2→
A US audit framework for confidentiality, integrity, availability, and privacy controls. For SaaS, the Type II audit (6–12 months of observation) is the standard enterprise baseline.
- ISO 27001→
International standard for Information Security Management Systems (ISMS). Often preferred in Europe instead of or alongside SOC 2. 3-year certification cycle.
- 0130 Sept 2026DField Q3 2026 roundup · what shifted, what we shipped, what is broken→
- 0201 Jul 2026DField Q2 2026 roundup · what shifted, what we shipped, what is broken→
- 0326 Apr 2026RAG's three failure modes · and the diagnostic table we use on every audit→
- 0426 Apr 2026We built our own LLM eval harness in 200 lines of TS · here is the file→
- 0526 Apr 2026Why your AI agent leaks money · 6 prompt-cache wins worth doing this week→
- 0626 Apr 2026OWASP LLM Top 10 v2 · what changed and what to ship→
- 0726 Apr 2026The CSP we ship · with notes on why each directive is there→
- 0823 Apr 2026On-device LLMs in 2026 · Gemini Nano vs Apple Intelligence for mobile builds→