Project 'The Log Remains' brief
- CISA and its partners published new agentic-AI security guidance on May 1, while SAG-AFTRA and the AMPTP reached a tentative successor deal on May 2. (cisa.gov) - The useful detail is the mismatch: one side is warning about autonomous systems in IT environments, while Hollywood is still negotiating consent rules around digital replicas. (cisa.gov) - That overlap makes provenance, approval logs, and human override feel less like paperwork and more like story material. (cisa.gov)
The interesting thing here is that the brief’s premise is not really speculative anymore. It sits right in the overlap between two live 2026 conversations — agen(cisa.gov)d fresh guidance on May 1 about how organizations should adopt agentic AI safely, and one day later SAG-AFTRA and the AMPTP announced a tentative successor TV/theatrical deal that still has to clear board review. (cisa.gov) So the core scenario — an AI assistant schedules and distributes a teaser using an arch(cisa.gov)e. One half is operational drift. The other is human permission. ### Why does the CISA piece matter? Because it treats agentic AI as a system that can take actions inside real workflows, not just generate text on command. The guide is about design, deployment, and operation, and it frames the risk as weak oversight around systems that can act across IT environments. That is basically the skeleton of your story problem — an assistant that doesn’t just suggest, but does. (cisa([cisa.gov)What kind of failure is this? It’s a provenance failure disguised as a creative mistake. The teaser is wrong because the system grabbed an archived voice model and an outdated story bible, but the deeper issue is traceability — who approved the asset, which version was authoritative, and what rules the system thought it was following. CISA’s guidance leans hard on oversight and alignment with existing risk frameworks, which maps neatly onto a narrative about missing logs and broken handoffs. (cisa.gov) ### Why bring in SAG-AFTRA? Because the voice-model part is not (cisa.gov) hard on digital replica protections, and the union’s March 20 statement on Val Kilmer’s digital replication made the baseline very plain — use of a deceased performer’s replica requires family consent under the union contract and state law, and the union said consent had been obtained in that case. (es.sagaftra.org) That gives your project a real-world pressure point. A reused voice is not neutral source materi(cisa.gov) the entertainment side is unresolved in exactly the right way. The AMPTP said on May 2 that the new SAG-AFTRA deal is tentative and that specific terms will not be released before board review. So the culture is actively negotiating the boundaries of AI use while security agencies are actively publishing controls for autonomous systems. (amptp.org)y happening. ### Why does the “outdated bible” detail work so well? Because stale context is how automation fails in ways humans instantly recognize. A bad model output is abstract. A teaser built from the wrong canon is legible. Everyone understands the horror of the system pulling from an old folder and treating it as truth. It turns governance into plot. ### Why these deliverables? They fit the problem. An interactive Twine or Ink scene can make the reader choose between speed, consent, and canon control. A short audio con(amptp.org)the project to expose its own provenance rules instead of just talking about them. ### What’s the bottom line? The brief works best if it stops treating AI as a spooky character and treats it as a permissions machine with bad defaults. That is where the real 2026 tension is — not “can the tool create,” but “who authorized the action, which source was live, and where is the log when something goes wrong.” (cisa.gov)
