Turn public exposure into a target profile the way an attacker now does it. Automated collection and aggregation at scale, language models reading a public writing sample for behavioural and communication-style signal, and computer vision pulling faces, locations and metadata out of an image set — with the legal line that runs through collection drawn as you go.

Lab Experience: Run an automated OSINT pipeline against a consented fictional persona set, then assemble a target profile you could defend in a report: what was collected, from where, under what authority, and what it actually supports.

Key Learning Areas:

• How AI changed the economics of reconnaissance — speed, cost and scale, not just quality
• Automated collection, aggregation and correlation across public and alternative sources
• Natural language processing for behavioural and communication-style profiling
• Computer vision for facial analysis, location inference and metadata extraction
• Where collection crosses a legal or authorisation boundary, and how to evidence that it did not

Build the attack. Synthetic video has moved past GAN face-swaps and long training runs to diffusion and few-shot synthesis; voice cloning now needs seconds of reference audio; language models write pretexts that read clean and hold a conversation. The day closes on real-time work — live avatars, virtual-camera injection, and attacks against remote identity verification.

Technical Mastery: Produce synthetic video, a cloned voice and a personalised campaign to a stated quality standard from consented reference material, then run a live pipeline inside a controlled call and record what it achieved and what it cost to build.

Key Learning Areas:

• From GANs to diffusion and one-shot synthesis: what genuinely changed, and what did not
• Few-shot facial synthesis, reenactment, and the honest quality ceiling of each pipeline
• Voice cloning, real-time conversion and the full vishing kill chain
• LLM pretexting, multi-turn conversation and campaign-scale personalisation
• Live avatars, virtual-camera injection and attacks on remote identity verification

Break what you built. Artefact, forensic and behavioural triage, then machine-learning detection deployed and measured against generators it has never seen — because benchmark scores hold steady while field performance quietly declines. The day ends on organisational defence, technical controls, threat intelligence, and a full-scale authorised engagement from scoping through to a report people act on.

Professional Applications: Attack your own detector and record where it fails, then design the out-of-band verification, payment-authority and incident-response controls that work without it — and run the final multi-vector simulation end to end.

Key Learning Areas:

• Artefact, forensic and behavioural detection, and precisely where each one stops working
• Deploying and evaluating ML detectors against unseen generators, codecs and compression
• Adversarial evasion, and how to read a detection vendor’s benchmark sceptically
• Provenance and content credentials, mail and call controls, injection-resistant identity assurance
• Threat intelligence, attribution, and scoping, authorising and reporting an AI-enhanced engagement