AI in Action: I/ITSEC 2025 Preview - Legal Hurdles, Rising Costs, and Tech Waffle in Military Training
AI in Action returns with a frank pre-I/ITSEC discussion led by host Andy Fawkes. Guests include Oliver Arup (Bohemia Interactive Simulations), Murat Kose (Quantum3D), and Paddy Little (Cervus Defence & Security). The focus: legal risk, spiralling costs, and vendor claims that look impressive in a slide deck but thin out under scrutiny.
If you work in legal, contracts, or compliance, this episode is worth your time. The short version: unclear IP rules, messy data rights, and inflated pricing models could slow procurement right as AI tools begin to add real value. Listen on Spotify.
What legal teams need to know before I/ITSEC
Oliver calls out "tech waffle": buzzwords that hide weak operational value and muddy legal obligations. Murat flags price and data access as key constraints. Paddy urges skepticism on AI features until the evidence and the licensing terms match the claims.
The legal takeaway is simple: the hard problems are data ownership, copyright, auditability, and total cost of use. If those aren't nailed down, projects stall-or worse, they launch with risk built in.
IP, data rights, and training data: set the rules early
- Data provenance: Is training data original, licensed, synthetic, or scraped? Who warrants non-infringement?
- Usage scope: Can vendors reuse your data or derived models to train other customers' systems?
- Outputs and copyright: Who owns generated terrains, behaviors, and scenarios? What about derivative works?
- Open-source obligations: Any copyleft code or datasets that could trigger reciprocal licensing?
- Retention and deletion: What happens to your data and fine-tuned models at contract end?
Costs that creep: price, compute, and lock-in
- Transparent pricing: Break out licenses, usage, compute, support, and data fees. Cap overages.
- Benchmarking rights: Compare vendor pricing to market without penalty.
- Exit ramps: Portable formats, model export, and transitional support to avoid lock-in.
- Performance-based payments: Tie fees to validated training outcomes, not demo promises.
Assurance for safety-critical training
- Model cards and logs: Require lineage, versioning, and audit logs for each model used in training.
- Testing and bias controls: Define acceptance criteria, red-team testing, and error budgets.
- Human oversight: Clear intervention points and operator authority for overrides.
- Security and export controls: Confirm ITAR/EAR handling, data residency, and incident reporting SLAs.
What the panel says is real (and useful)
Beneath the hype, practical gains are showing up: predictive analytics for readiness, adaptive learning that tunes scenarios to skill level, and generative tools that speed terrain and behavior modeling. These help-but only with clean data rights, measurable outcomes, and oversight baked into contracts.
What exhibitors plan to bring
- BISim: Next-gen simulation platforms and AI-assisted content generation.
- Quantum3D: UAV simulators and high-fidelity dome projection systems.
- Cervus: Advanced analytics to evaluate training effectiveness and decision-making.
All worth seeing-just pair the product tour with pointed legal and procurement questions.
Questions to ask on the show floor
- What warranties cover training data provenance and non-infringement of outputs?
- Can we review model cards, validation reports, and change logs for each release?
- Which datasets and foundation models do you rely on, and how are they licensed?
- What is the total cost of ownership across 3-5 years, including compute and support?
- Do we retain ownership of fine-tuned models and generated content?
- How do you handle export controls, data segregation, and incident response?
- What measurable outcomes will you commit to, and how do we verify them?
Helpful references
Upskill your team
If your legal or procurement teams are expanding their AI literacy, explore practical training options that map to job roles and workflows.
AI will keep moving into defence training. Your edge is crisp contracts, provable assurance, and a budget that pays for outcomes-not buzzwords. If you're headed to Orlando from December 1-4, bring this checklist and make the conversations count.
Your membership also unlocks:
