Truth, Provenance and Synthetic Evidence
A repeatable toolkit for verifying what you see — and the authentication problem that synthetic media has just handed the profession.
The hook
“You are going to be worse at spotting a fake than you think you are — the room will prove it in four minutes. Now imagine the artefact you just misjudged was tendered as an exhibit, and you were the one who had to challenge it.”
What you'll be able to do
- Apply the SIFT method — Stop, Investigate the source, Find better coverage, Trace the claim to its origin — to a viral claim and to an AI's answer alike.
- Practise lateral reading as a habit: leave the page and check the source in new tabs.
- Explain the confidence gap in detecting synthetic media, and the liar's dividend that follows from it.
- Frame authentication of an electronic record as the legal problem it is — certificate, provenance, chain of custody — rather than as a technology novelty.
On the syllabus
- The SIFT method — Stop, Investigate the source, Find better coverage, Trace the claim to its origin
- Lateral reading as a verification habit, and the emotional spike as a signal to slow down
- Deepfakes and synthetic media: how they are made and why they persuade
- The confidence gap, and the liar's dividend that follows from it
- Authentication as doctrine: electronic records, the certificate regime, provenance and chain of custody
In short
Session 5 turns the decoding skills of Sessions 1 and 2 into a repeatable verification routine, and then puts that routine to work on the problem synthetic media has created for the profession. SIFT gives four moves that work identically on a forwarded message, a news story, a deepfake and a chatbot's answer; lateral reading is the habit that makes them fast. The room takes a real-or-AI test and discovers its own confidence gap. The hour then turns legal: if provenance is the question, authentication is the doctrine, and the certificate regime for electronic records is where the argument will actually be had. The hands-on half hour runs a SIFT drill and ends with a short authentication note.
Why it matters for using AI well
Never rely on, share, or tender what you have not traced to its origin. Stop, investigate, find, trace — and treat the strength of your own reaction, or the model's confidence, as the cue to check rather than the reason not to. When the artefact matters, provenance is the argument.
What they leave with
The skill
Run SIFT to origin on anything you would rely on or share, and state what would be needed to authenticate an electronic record.
The insight
Your confidence in your own eyes is not evidence — and in a world of cheap fakes, provenance, not plausibility, is what a claim rests on.
The moment they remember
The real-or-AI test, run as a commit-then-reveal on their phones: image by image, each student commits to a verdict and to how confident they are. The tally shows a room that scored far below what it predicted — and it is public, collective and impossible to argue with. The turn that makes it land is the next slide: the same artefact, tendered as an exhibit. The humbling becomes professional in a single step, and the method they are then handed feels like equipment rather than advice.
In this session
- 01
Stop: a strong emotional reaction — outrage, delight, vindication — is the moment your judgment is weakest and the moment you are most likely to share. The habit is to notice the spike and pause on it.
- 02
Investigate the source: who is telling me this, how do they know, and what are they selling? (Bergstrom & West.) Lateral reading — leaving the page to see what others say about the source — is faster and more reliable than reading the page harder.
- 03
Find better coverage and trace to the original: get to the primary document, the original quote, the actual study, the actual judgment. An AI's summary of a source is never a substitute for the source, and a screenshot is not a document.
- 04
Deepfakes and synthetic media: how they are made, why they are persuasive, and why detection tools are a moving target rather than an answer. For a lawyer this is an authentication problem, not a novelty.
- 05
The confidence gap: people are markedly worse at spotting fakes than they believe, and the gap is widest exactly where the stakes are highest. The same over-confidence is what makes a fluent chatbot answer feel true.
- 06
The liar's dividend (Chesney & Citron): once everyone knows that convincing fakes exist, genuine evidence can be dismissed as fake. The corrosive effect runs in both directions, and it lands on courts before it lands anywhere else.
- 07
Authentication as doctrine: an electronic record has to be got into evidence, and India's regime turns on a certificate — s.65B of the Indian Evidence Act, 1872, now carried into s.63 of the Bharatiya Sakshya Adhiniyam, 2023, which came into force on 1 July 2024. Two Supreme Court decisions set the position under the old section and carry over: Anvar P.V. v. P.K. Basheer (2014) 10 SCC 473 held the certificate mandatory, overruling Navjot Sandhu; and Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal (2020) 7 SCC 1 settled the position after conflicting authority, overruling Shafhi Mohammad — holding the certificate mandatory for secondary evidence of an electronic record, but not required where the original is itself produced.
- 08
What the BSA changed, and why it matters for synthetic media: under the old s.65B(4) the certificate was signed by the person responsible for the device. Section 63(4)(c) of the BSA requires the certificate to be signed by the person in charge of the computer or communication device — or of the management of the relevant activities — *and* an expert, in the format prescribed by the Schedule. The legislature has, in other words, put a technical expert into the chain of every contested electronic record, which is precisely the pressure point synthetic media creates. In practice the question a court asks is not “is this a deepfake?” but “where did this file come from, who can speak to that, and what does the hash, the metadata and the chain of custody show?”
- 09
SIFT on the machine: ask a model for sources on a niche claim and it produces plausible-looking references. Run SIFT on them live and many do not survive — which is the bridge to the authority check drilled in Session 6.
The four-step mirror
Run it on the class. Then on the machine.
An experiment on the room, the same effect explained in the model, a live demonstration on a real chatbot, and a named takeaway skill.
Real or AI? — and the confidence gap
On the class
A sequence of images and clips: the room commits, on their phones, to real or AI for each, and to how confident they are. The tally reveals the room scored worse than it predicted.
In the model
Generative models produce synthetic media that is fluent and convincing yet tethered to no real event — the same way a language model produces confident text untethered from truth.
Live chatbot
The answers are revealed; the confidence gap goes on screen; and the presenter re-frames the best-scoring fake as a disputed exhibit, asking the room what they would now need to authenticate it.
The skill
Your confidence in a judgment is not evidence that it is right. Demand provenance in proportion to the stakes.
Would you share this?
On the class
An outrage headline goes up; the room votes, fast, on whether they would forward it.
In the model
Outrage spreads faster than truth, and the same incentives shaped the corpus the model learned from.
Live chatbot
The presenter pastes the headline into a chatbot and asks “is this true?” — it confidently elaborates. Then the class runs SIFT on the same claim together — stop, investigate, find, trace — and reaches the origin the model never checked.
The skill
Emotional spike means stop. Verify before you share, and before you rely.
Make it cite, then check
On the class
Vote: how confident are you that the sources an AI gives for a claim are real?
In the model
Asked for references on a niche claim, a model composes plausible-looking citations — authors, journals, years — with the same mechanism it uses to invent a case.
Live chatbot
The presenter asks a chatbot for sources on a niche claim and runs SIFT on each, live. Some exist and say something different; some do not exist at all.
The skill
Make the AI cite, then verify that the citation exists and says what the AI says it says.
Hands-on · on your own laptop
The SIFT Drill and the Authentication Note
Two verification runs and one short piece of legal writing. Trace a viral legal claim to its origin with SIFT; run SIFT on three sources a chatbot supplies for a proposition in your field; then write the five-line authentication note you would need if the first artefact were tendered as evidence.
Run of show · 30 minutes
- 0–10 min — Take a viral legal-news claim (a forwarded “the Supreme Court has just held…”) and run all four SIFT moves until you reach the origin or establish that there isn't one.
- 10–20 min — Ask a chatbot for three sources supporting a proposition in your field and run SIFT on each. Record a verdict per source: verified, corrected, or fabricated.
- 20–28 min — Write five lines: if this item were tendered as evidence, what would be needed to authenticate it, and what would your challenge be?
- 28–30 min — Verdicts are tallied across the room; the aggregate survival rate of AI-supplied sources goes on screen.
Deliverable
A two-part verification log — one viral claim, three AI-supplied sources — with the SIFT steps and a verdict for each, plus a five-line authentication note framing what provenance the artefact would require.
Key sources & cases
Mike Caulfield, SIFT / Web Literacy for Student Fact-Checkers
The four moves — Stop, Investigate the source, Find better coverage, Trace to the original — and lateral reading (CC BY 4.0).
Bergstrom & West, Calling Bullshit (2020)
Who's telling me this? How do they know? What are they selling? — the triage questions for any confident source.
Chesney & Citron, “Deep Fakes” (California Law Review, 2019)
The legal analysis of synthetic media and the source of the “liar's dividend.”
Sam Gregory / WITNESS — guidance on deepfakes
Practical guidance on synthetic media and its misuse from the human-rights video organisation.
MIT Media Lab — Detect Fakes
The real-or-AI test used to surface the room's confidence gap.
Bharatiya Sakshya Adhiniyam, 2023, s.63 (successor to Indian Evidence Act, 1872, s.65B)
The admissibility regime for electronic records. Sub-section (4)(c) requires a certificate signed by the person in charge of the device or the relevant activities *and* an expert, in the Schedule's format — a change from s.65B(4), which required only the responsible person. In force 1 July 2024. Verified 2026-08-26 against the section text on Indian Kanoon; confirm against the Gazette before academic-council use.
Anvar P.V. v. P.K. Basheer (2014) 10 SCC 473
Three-judge bench (Lodha CJI, Kurian Joseph and Nariman JJ), 18 September 2014: compliance with s.65B is mandatory for secondary electronic evidence, overruling State v. Navjot Sandhu on this point. Arose from an election petition where campaign CDs were tendered without certification. Verified 2026-08-26.
Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal (2020) 7 SCC 1
Three-judge bench, 14 July 2020: the s.65B(4) certificate is mandatory for *secondary* evidence of an electronic record but is not required where the original is produced; Shafhi Mohammad (2018) 2 SCC 801 overruled. Verified 2026-08-26. Teach the primary/secondary distinction — it is the part most often misstated.
Readings
- Mike Caulfield, SIFT / Web Literacy for Student Fact-Checkers
- Bergstrom & West, Calling Bullshit (2020)
- Chesney & Citron, “Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security” (2019)
- WITNESS — guidance on deepfakes and synthetic media
- Bharatiya Sakshya Adhiniyam, 2023, s.63; Anvar P.V. v. P.K. Basheer (2014) 10 SCC 473; Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal (2020) 7 SCC 1
Next session
Session 06 / 08
Grounding the Machine — Retrieval, Tools and the Authority Check
Ground it and check itSixteen hours, one professional discipline.
Using AI well is not a knack — it is competence, candour and verification, practised until they are habits you could defend in court.