imadethisup.org Retain →
BLOG

Short, sourced notes on a moving field.

Case analyses, research summaries, and policy explainers from the editorial team. Every post links to the same primary sources we cite elsewhere on the site. Available as RSS.

31 of 31 posts, newest first.

  1. 2026-08-30
    What invisible watermarks can and cannot survive. Research on attacking invisible watermarks reveals removal, forgery and detection-degradation are distinct threats. Robustness is conditional, not absolute.
    PROVENANCE
    Read →
  2. 2026-08-30
    Reporting non-consensual imagery: what one audit found. An audit of reporting mechanisms for non-consensual intimate imagery shows different outcomes depending on which platform route victims use.
    PERSONAL
    Read →
  3. 2026-08-29
    Why deepfake impersonation works, and what stops it. Deepfake attacks on executives target decision-making, not media detection. The control that works is procedural verification outside the attack channel.
    BUSINESS
    Read →
  4. 2026-08-29
    When audio deepfake detectors fail on new voice generators. Audio deepfake detectors trained on one set of voice generators often fail on others. What the research says about using detection results in reports.
    RESEARCH
    Read →
  5. 2026-08-29
    What the US AI Safety Institute says still needs research. The US AI Safety Institute has published specific research gaps in synthetic content mitigation. Here is what it says does not work yet.
    RESEARCH
    Read →
  6. 2026-08-29
    Penalties for breaching the EU AI Act's disclosure rules. Article 99(4)(g) puts breaches of the AI Act's synthetic-content transparency duties in the EUR 15 million or 3 percent tier. Who enforces, and how fines are set.
    LAW
    Read →
  7. 2026-08-29
    What Content Credentials establish, and what they do not. Content Credentials are signed statements about a file's history, not proof that the depicted events happened. Their absence is uninformative on its own.
    PROVENANCE
    Read →
  8. 2026-08-29
    Detector results on deepfakes found in the wild. Deepfake-Eval-2024 tested detectors on deepfakes actually circulating online. Reported AUC fell by 45 to 50 percent against laboratory benchmarks.
    RESEARCH
    Read →
  9. 2026-08-29
    Synthetic media disclosure obligations now in force. Providers and deployers must mark and disclose AI-generated content under Article 50. Requirements took effect 2 August 2026.
    LAW
    Read →
  10. 2026-08-29
    What the C2PA conformance registry actually lists. The C2PA registry lists seven devices with certified Content Credentials. Which camera makers appear, which haven't, and how you can verify it yourself.
    PROVENANCE
    Read →
  11. 2026-08-29
    What a deepfake detector score does not tell you. Detectors score highly on the benchmark they trained on. Measured performance on generators released later is a different and much weaker number.
    RESEARCH
    Read →
  12. 2026-06-15
    Reg S-P’s new baseline for smaller firms. The SEC's amended Reg S-P turns a dusty privacy rule into a cybersecurity mandate, written incident response, 30-day customer breach notice, and vendor oversight, with the smaller-firm deadline now here.
    BUSINESSLAW
    Read →
  13. 2026-06-15
    Defending AI-assisted e-discovery review. TAR has been judicially blessed for over a decade and generative-AI review is joining the workflow, but the hard part is defending the process: proportionality, validation by recall and precision, and a human who certifies under Rule 26(g).
    LAWRESEARCH
    Read →
  14. 2026-06-15
    AI governance: what the CISO actually owns. When an organization adopts AI, governance lands on the security chair, the CISO's remit across NIST AI RMF, ISO/IEC 42001, the EU AI Act, deepfake risk, and board reporting.
    RESEARCHBUSINESS
    Read →
  15. 2026-06-15
    Arbitration’s new rules for AI evidence. The JAMS AI Disputes Rules show how to gate AI-touched evidence, secured-environment expert access to models and data, disclosure of AI use, and proportionality.
    LAWRESEARCH
    Read →
  16. 2026-06-15
    AI tools and attorney-client privilege. Pasting client material into a third-party AI tool can breach confidentiality and waive privilege, what ABA Rule 1.6, Formal Opinion 512, and FRE 502 require counsel to do.
    LAWBUSINESS
    Read →
  17. 2026-06-15
    A deepfake incident-response plan for small firms. A checklist-driven playbook for voice-clone and deepfake incidents, detect, contain, preserve, notify, recover, grounded in NIST SP 800-61, FBI IC3, and CISA guidance.
    BUSINESSCASE
    Read →
  18. 2026-06-15
    Agentic AI: who answers when an agent trades?. Autonomous AI agents are trading and drafting disclosures with little supervision, so when one breaks the securities laws, who answers? The SEC's "AI washing" cases and why agency law keeps the firm liable.
    RESEARCHLAW
    Read →
  19. 2026-06-15
    The hallucination tax on unverified AI citations. A fabricated AI citation is no glitch; it's a self-inflicted breach of the duty of competence that taxes the whole proceeding in sanctions, wasted hours, and lost credibility.
    LAWRESEARCH
    Read →
  20. 2026-06-15
    “Follow the money” is no longer enough. A cloned voice induces the wire and on-chain layering erases the trail, why financial tracing alone no longer proves modern fraud, and what investigators must add.
    LAWBUSINESS
    Read →
  21. 2026-06-15
    Who decides when digital evidence is contested?. When adversaries deadlock over whether digital or AI evidence is authentic, courts can appoint a neutral technical authority under FRCP 53 or FRE 706 whose only client is the record.
    LAWRESEARCH
    Read →
  22. 2026-06-15
    Investigating fraud when the evidence can be fake. Generative AI lets fraudsters fabricate documents and voices, and lets investigators be misled by fluent synthetic artifacts. How curiosity and rigor keep an investigation honest.
    CASERESEARCH
    Read →
  23. 2026-06-15
    Deepfake wire fraud in real-estate closings. Voice and video impersonation is hitting title, escrow, and closing workflows. The break point isn't the model; it's the missing out-of-band callback on wiring instructions.
    CASEBUSINESS
    Read →
  24. 2026-06-15
    Deepfakes and evidence integrity in family court. In custody court a fabricated clip lands before anyone proves it fake, but Rules 901 and 902, used affirmatively, put the authentication burden back where it belongs.
    LAWRESEARCH
    Read →
  25. 2026-06-15
    Authenticating AI evidence: is Rule 901 enough?. Rule 901's lenient gate met its match in generative AI, 901(b)(9), 902(13)-(14), proposed Rule 707, and what litigators should do now.
    LAWACADEMIC
    Read →
  26. 2026-06-15
    Deepfakes are reshaping workplace harassment. Synthetic intimate imagery of a coworker is a workplace event, not a private one, and Title VII reaches conduct that never happened at the office.
    LAWPERSONAL
    Read →
  27. 2026-06-15
    When the file lies about itself. Generative AI can fabricate the hidden author, date, and provenance metadata we rely on to prove a file is authentic. How to verify with hashing and Content Credentials.
    RESEARCHPROVENANCE
    Read →
  28. 2026-06-15
    The copyright fight over generative-AI creations. Two copyright fights collide over generative AI, whether AI output can be owned, and whether training on copyrighted works was lawful.
    LAWRESEARCH
    Read →
  29. 2026-04-27
    What the Arup deepfake actually proves. The break point wasn't the model. It was the absence of an out-of-band callback above the wire-transfer threshold.
    CASEBUSINESS
    Read →
  30. 2026-04-27
    Why deepfake detectors fail on new generators. A short tour of the cross-generator generalization gap, the field's central open problem.
    RESEARCHACADEMIC
    Read →
  31. 2026-04-27
    What the TAKE IT DOWN Act changes. A practical reading of S.146 for survivors, platforms, and counsel.
    LAWPERSONAL
    Read →

More posts coming. Suggest a topic at info@imadethisup.org.