Principal Engineer, AI Security, Model Internals

Opera

. Build tooling to inspect open-weight models layer by layer and head by head .

Posted 11/9/2026full-timeWrocław • PolandLeadWebsite

Core Competencies

Role fit

Core Competencies

Use this summary to align your resume positioning with the role.

Demonstrates strong engineering expertise in Python and PyTorch, with practical experience in building secure fine-tuning pipelines and evaluating trained models. Capable of translating research into actionable prototypes while maintaining a focus on AI security and adversarial resilience.

Highest-signal resume keywords

Python ProgrammingPyTorch FrameworkTransformer InternalsFine-Tuning ExperienceAdversarial Mindset

ATS Keywords

Tailor your resume

Applicant Tracking System Keywords

Tip: use these terms in your resume and cover letter to boost ATS matches.

Hard Skills

Model EvaluationResidual Stream ExplanationModel Forward Pass HookingLoRA Fine-TuningQLoRA Fine-TuningPEFT Fine-TuningRed-TeamingVulnerability ResearchReverse EngineeringCTF Experience

Soft Skills

Clear Written EnglishConfidence to Challenge Ideas

Tools & Technologies

TransformerLensNnsightActivation PatchingAblationSparse Autoencoders

Industry Keywords

AI SecurityModel Internal State InspectionData Poisoning ResistanceSafety RegressionsOn-Device Inference

Tech Stack

Tools & technologies

PythonPyTorch

About the role

Key responsibilities & impact

Build tooling to inspect open-weight models layer by layer and head by head
Capture activations and use linear probes, logit and tuned lens, activation patching, ablation, and sparse autoencoders
Scan third-party weights and fine-tunes for backdoors, trojans, and hidden behaviour
Design runtime monitors for agents that inspect model internal state during critical moments such as tool calls
Escalate suspicious behaviour to a second model for verification
Red-team Opera’s agentic browsing and develop prompt-injection defenses
Design architectures that keep untrusted web content away from privileged actions
Build secure LoRA and other fine-tuning pipelines resistant to data poisoning
Test models for memorisation, leakage, and safety regressions
Reproduce promising research papers and turn validated techniques into monitors and checks for on-device inference
Document findings, open-source tools where appropriate, and help build Opera’s voice in AI security
Work closely with Opera’s AI leadership as a technical partner evaluating and challenging engineering approaches

Requirements

What you’ll need

Strong engineering experience in Python and PyTorch
Track record of building real systems, not only notebooks
Hands-on experience with transformer internals
Ability to explain the residual stream and hook into a model’s forward pass
Experience with TransformerLens, nnsight, or raw hooks
Practical fine-tuning experience with LoRA, QLoRA, and PEFT
Experience evaluating trained models
Evidence of an adversarial mindset through red-teaming, vulnerability research, reverse engineering, CTFs, jailbreak work, or comparable activities
Ability to take an idea from paper to working prototype quickly
Ability to distinguish real findings from artefacts
Clear written English
Confidence to challenge ideas, including a manager’s
CV must be submitted in English

Benefits

Comp & perks

Diverse and inclusive team environment
Flat structure with short decision-making paths
Ongoing skill development
Friendly and empowering work environment
Remote work options are addressed through the company FAQ