Internship: AI Safety, Jailbreaking Attacks & Defense, Agentic AI, Human Behavior(Ph.D. Research Intern)Company: Centific
Location: Seattle, WA (or Remote)
Type: Full-time Internship - 40 hours per week
Build the Future of Safe and Responsible AIAre you advancing the frontiers of AI safety, LLM jailbreak detection and defense, and agentic AI-with publications to show for it? Join us to translate pioneering research into robust security and trustworthy LLM systems that resist adversarial and behavioral exploits.
The MissionWe're tackling cutting-edge AI safety across adversarial robustness, jailbreak defense, agentic workflows, and human-in-the-loop risk modeling. As a Ph.D. Research Intern, you'll own high-impact experiments from concept to prototype to deployable modules, directly contributing to our platform's security guarantees.
What You'll Do- Advance AI Safety: Design, implement, and evaluate attack and defense strategies for LLM jailbreaks (prompt injection, obfuscation, narrative red teaming).
- Evaluate AI Behavior: Analyze and simulate human-AI interaction patterns to uncover behavioral vulnerabilities, social engineering risks, and over-defensive vs. permissive response tradeoffs.
- Agentic AI Security: Prototype workflows for multi-agent safety (e.g., agent self-checks, regulatory compliance, defense chains) that span perception, reasoning, and action.
- Benchmark & Harden LLMs: Create reproducible evaluation protocols/KPIs for safety, over-defensiveness, adversarial resilience, and defense effectiveness across diverse models (including latest benchmarks and real-world exploit scenarios).
- Deploy and Monitor: Package research into robust, monitorable AI services using modern stacks (Kubernetes, Docker, Ray, FastAPI); integrate safety telemetry, anomaly detection, and continuous red-teaming.
Example Problems You Might Tackle- Jailbreaking Analysis: Systematically red-team advanced LLMs (GPT-4o, GPT-5, LLaMA, Mistral, Gemma, etc.), uncovering novel exploits and defense gaps.
- Multi-turn Obfuscation Defense: Implement context-aware, multi-turn attack detection and guardrail mechanisms, including countermeasures for obfuscated prompts (e.g., StringJoin, narrative exploits).
- Agent Self-Regulation: Develop agentic architectures for autonomous self-check and self-correct, minimizing risk in complex, multi-agent environments.
- Human-Centered Safety: Study human behavior models in adversarial contexts-how users probe, trick, or manipulate LLMs, and how defenses can adapt without excessive over-defensiveness.
Minimum Qualifications- Ph.D. student in CS/EE/ML/Security (or related); actively publishing in AI Safety, NLP robustness, or adversarial ML (ACL, NeurIPS, BlackHat, IEEE S&P, etc.).
- Strong Python and PyTorch/JAX skills; comfort with toolkits for language models, benchmarking, and simulation.
- Demonstrated research in at least one of: LLM jailbreak attacks/defense, agentic AI safety, human-AI interaction vulnerabilities.
- Proven ability to go from concept 12 code 12 experiment 12 result, with rigorous tracking and ablation studies.
Preferred Qualifications- Experience in adversarial prompt engineering, jailbreak detection (narrative, obfuscated, sequential attacks).
- Prior work on multi-agent architectures or robust defense strategies for LLMs.
- Familiarity with red-teaming, synthetic behavioral data, and regulatory safety standards.
- Scalable training and deployment: Ray, distributed evaluation, CI/telemetry for defense protocols.
- Public code artifacts (GitHub) and first-author publications or strong open-source impact.
Our Stack (you'll touch a subset)- Modeling: PyTorch/JAX, Hugging Face, OpenMMLab, Mistral, LLaMA
- Safety: Red-teaming frameworks, LLM benchmarking (SODE, ART), human behavior simulation
- Systems: Python, Ray, Kubernetes, Docker, FastAPI, Triton, Weights & Biases
- Defense Pipelines: Context-aware filtering, prompt manipulation detection, anomaly telemetry
What Success Looks Like- A publishable outcome (with company approval) or production-ready module measurably improving safety KPIs: adversarial robustness, over-defensiveness, and incident response latency.
- Clean, reproducible code with documented ablations and end-to-end rerun reports for safety benchmarks.
- A demo that communicates capabilities, limits, and next steps in defense and security assurance.
Why Centific- Real Impact: Your research ships-directly securing our core features and AI infrastructure.
- Mentorship: Collaborate with Principal Architects and senior researchers in AI safety and adversarial ML.
- Velocity + Rigor: Balance high-quality research with mission-critical product focus.
Rate: $50 per hour 
