# IEEE Spectrum: Hany Farid on Trusting Visual Recordings Again

[Canonical reading](https://escapemeridian.com/readings/ieee-spectrum-hany-farid-on-trusting-visual-recordings-again-a8e9fabe)

IEEE Spectrum’s Hany Farid argues that generative AI calls for a “presumption of fake”: creators, rather than skeptics, should bear the burden of showing that recordings are real. He recounts examining a video that appeared to show a Tomahawk missile hitting a school. After assessing its motion, apparent size and the delay between impact and sound, he judged it authentic. But he says the investigation also showed how suspicion can persist even when signs of manipulation are absent.

Farid argues that forensic analysis cannot keep pace with manipulation’s scale and speed. He says provenance signals can help legitimate creators establish authenticity, even though they need not be adversary-proof: their purpose is to let good actors show what their recordings are, not simply catch bad actors. He extends the proposal to video calls, where real-time deepfakes leave little time for frame-by-frame human analysis. Video-conferencing software could check whether frames were cryptographically signed by a physical camera sensor or altered in transit. Farid presents provenance and forensic analysis as complementary: provenance can support trust in legitimate recordings, while forensics can help identify manipulation. Whether enough good actors will adopt provenance signals to begin drowning out bad actors remains conditional in his account.

Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.

## Original source

IEEE Spectrum · Article

Original source date: Nov 20, 2026 (2026-11-20T14:00:03+00:00).

Date basis: Date stated by the publisher.

Added to Escape Meridian: 2026-11-20T18:46:46.655Z. First observed: 2026-11-20T14:28:13.459Z. Retrieved: 2026-11-20T18:45:53.286Z.

- [I Don’t Trust Visual Recordings Anymore. But One Day I Will Again](https://spectrum.ieee.org/fighting-back-against-deep-fakes)

## Limits

Farid’s proposal that enough good actors adopting provenance signals could begin to drown out bad actors is conditional; the article does not specify what level of adoption would be enough. The article presents Farid’s account of his analysis of the missile video but does not provide the video or the underlying analysis for independent assessment.


## Open question

What level of adoption of provenance signals would be enough for good actors to begin drowning out bad actors, as Farid proposes?


## Source-stated

Farid argues that the burden of proof should shift to creators to show their media is real, a change he says reflects the conditions created by generative AI.

> the burden is now on the creator to prove that their content is real.

- [Supporting source](https://spectrum.ieee.org/fighting-back-against-deep-fakes)

## Source-stated

Farid says his analysis of the missile video led him to judge it authentic, while also revealing how hard it had become for him to trust a recording.

> I finally decided that it was authentic, and I gave the reporters my updated view.

- [Supporting source](https://spectrum.ieee.org/fighting-back-against-deep-fakes)

## Source-stated

Farid argues provenance signals can help legitimate creators establish authenticity, even if the signals are not adversary-proof.

> Their purpose is to let good actors prove that their audiovisual recording is what it claims to be.

- [Supporting source](https://spectrum.ieee.org/fighting-back-against-deep-fakes)

## Source-stated

Farid proposes that video-conferencing software could check whether live video frames were signed by a physical camera sensor or altered in transit.

> the software could verify in real time whether the frames it’s receiving were signed by a physical camera sensor or whether they were altered in transit.

- [Supporting source](https://spectrum.ieee.org/fighting-back-against-deep-fakes)

## Source-stated

Farid argues that provenance signals and forensic analysis should work together, with provenance supporting trust in legitimate recordings and forensics helping identify manipulation.

> The two approaches work together, with provenance allowing good actors to be trusted, while forensic analyses root out the bad actors.

- [Supporting source](https://spectrum.ieee.org/fighting-back-against-deep-fakes)

## Source and contributor context

Hany Farid · Article author

These are credits stated by the source; they do not establish a profile match or authorship of every passage.


## Source media

Media declared by the original publisher; availability checked separately from the editorial reading.

- [image · IEEE Spectrum](https://spectrum.ieee.org/media-library/photo-collage-of-a-mans-face-featuring-glitched-distortions-and-abstract-shapes.jpg?id=68721975&width=1200&height=600&coordinates=143%2C771%2C114%2C609)

## Terms and references

Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.


## IEEE Spectrum

IEEE Spectrum reports on engineering practice, research, and the industries around it.

- [Explore this context](/entities/ieee-spectrum)
- [Reference](https://spectrum.ieee.org/)
- [Reference](https://spectrum.ieee.org/feeds/feed.rss)

## TIME

TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.

- [Explore this context](/entities/time)
- [Reference](https://time.com/)

## Editorial provenance

Version 2. Publication ID: 637a0a30-8032-4886-8468-4aa7870cff93.

Prepared by gpt-6-luna; grounding review by gpt-6-luna.

Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-20T18:46:46.655Z. This checks fidelity to the source, not whether its claims are independently true.

Editorial selection: accept; reasons: useful\_perspective.

Observation: 300a87487c475e6034714dc1e796c4f2026d7e80719f17bb2788682b6bb0561b. Source text hash: 8dbfe5391d34ab93590c52e34d3ca8c37c416c40133bf68feccfb459162c1db8.

Correction reason: Revised the summary to attribute the proposal and its conditional adoption claim more clearly to Farid, while retaining the prior claims, question and limitations.


## Revision and source-check history

### Version 2 — 2026-11-20T18:46:46.655Z

IEEE Spectrum’s Hany Farid argues that generative AI calls for a “presumption of fake”: creators, rather than skeptics, should bear the burden of showing that recordings are real. He recounts examining a video that appeared to show a Tomahawk missile hitting a school. After assessing its motion, apparent size and the delay between impact and sound, he judged it authentic. But he says the investigation also showed how suspicion can persist even when signs of manipulation are absent.

Farid argues that forensic analysis cannot keep pace with manipulation’s scale and speed. He says provenance signals can help legitimate creators establish authenticity, even though they need not be adversary-proof: their purpose is to let good actors show what their recordings are, not simply catch bad actors. He extends the proposal to video calls, where real-time deepfakes leave little time for frame-by-frame human analysis. Video-conferencing software could check whether frames were cryptographically signed by a physical camera sensor or altered in transit. Farid presents provenance and forensic analysis as complementary: provenance can support trust in legitimate recordings, while forensics can help identify manipulation. Whether enough good actors will adopt provenance signals to begin drowning out bad actors remains conditional in his account.

## Original source

IEEE Spectrum · Article

Original source date: Nov 20, 2026 (2026-11-20T14:00:03+00:00).

Date basis: Date stated by the publisher.

Added to Escape Meridian: 2026-11-20T18:46:46.655Z. First observed: 2026-11-20T14:28:13.459Z. Retrieved: 2026-11-20T18:45:53.286Z.

- [I Don’t Trust Visual Recordings Anymore. But One Day I Will Again](https://spectrum.ieee.org/fighting-back-against-deep-fakes)

## Limits

Farid’s proposal that enough good actors adopting provenance signals could begin to drown out bad actors is conditional; the article does not specify what level of adoption would be enough. The article presents Farid’s account of his analysis of the missile video but does not provide the video or the underlying analysis for independent assessment.


## Open question

What level of adoption of provenance signals would be enough for good actors to begin drowning out bad actors, as Farid proposes?


## Source-stated

Farid argues that the burden of proof should shift to creators to show their media is real, a change he says reflects the conditions created by generative AI.

> the burden is now on the creator to prove that their content is real.

- [Supporting source](https://spectrum.ieee.org/fighting-back-against-deep-fakes)

## Source-stated

Farid says his analysis of the missile video led him to judge it authentic, while also revealing how hard it had become for him to trust a recording.

> I finally decided that it was authentic, and I gave the reporters my updated view.

- [Supporting source](https://spectrum.ieee.org/fighting-back-against-deep-fakes)

## Source-stated

Farid argues provenance signals can help legitimate creators establish authenticity, even if the signals are not adversary-proof.

> Their purpose is to let good actors prove that their audiovisual recording is what it claims to be.

- [Supporting source](https://spectrum.ieee.org/fighting-back-against-deep-fakes)

## Source-stated

Farid proposes that video-conferencing software could check whether live video frames were signed by a physical camera sensor or altered in transit.

> the software could verify in real time whether the frames it’s receiving were signed by a physical camera sensor or whether they were altered in transit.

- [Supporting source](https://spectrum.ieee.org/fighting-back-against-deep-fakes)

## Source-stated

Farid argues that provenance signals and forensic analysis should work together, with provenance supporting trust in legitimate recordings and forensics helping identify manipulation.

> The two approaches work together, with provenance allowing good actors to be trusted, while forensic analyses root out the bad actors.

- [Supporting source](https://spectrum.ieee.org/fighting-back-against-deep-fakes)

## Source and contributor context

Hany Farid · Article author

These are credits stated by the source; they do not establish a profile match or authorship of every passage.


## Source media

Media declared by the original publisher; availability checked separately from the editorial reading.

- [image · IEEE Spectrum](https://spectrum.ieee.org/media-library/photo-collage-of-a-mans-face-featuring-glitched-distortions-and-abstract-shapes.jpg?id=68721975&width=1200&height=600&coordinates=143%2C771%2C114%2C609)

## Terms and references

Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.


## IEEE Spectrum

IEEE Spectrum reports on engineering practice, research, and the industries around it.

- [Explore this context](/entities/ieee-spectrum)
- [Reference](https://spectrum.ieee.org/)
- [Reference](https://spectrum.ieee.org/feeds/feed.rss)

## TIME

TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.

- [Explore this context](/entities/time)
- [Reference](https://time.com/)

## Editorial provenance

Version 2. Publication ID: 637a0a30-8032-4886-8468-4aa7870cff93.

Prepared by gpt-6-luna; grounding review by gpt-6-luna.

Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-20T18:46:46.655Z. This checks fidelity to the source, not whether its claims are independently true.

Editorial selection: accept; reasons: useful\_perspective.

Observation: 300a87487c475e6034714dc1e796c4f2026d7e80719f17bb2788682b6bb0561b. Source text hash: 8dbfe5391d34ab93590c52e34d3ca8c37c416c40133bf68feccfb459162c1db8.

Correction reason: Revised the summary to attribute the proposal and its conditional adoption claim more clearly to Farid, while retaining the prior claims, question and limitations.


### Version 1 — 2026-11-20T14:32:53.312Z

IEEE Spectrum’s Hany Farid argues that generative AI has shifted the burden of trust in visual media: creators should demonstrate that recordings are real, rather than skeptics having to establish that they are fake. He describes investigating a video that appeared to show a Tomahawk missile hitting a school. After checking the missile’s path, apparent size and the delay between impact and sound, he judged the video authentic. Yet the experience showed him how suspicion can persist even when signs of manipulation are absent.
Farid argues that forensic analysis alone cannot keep pace with the scale and speed of manipulation. He proposes combining it with provenance signals, such as cryptographically signed records of a recording’s origin and edits, so legitimate creators can establish authenticity. He says these signals need not be impossible to attack: their purpose is to help good actors show what their recordings are, not simply catch bad actors.
He extends the argument to video calls, where real-time deepfakes can impersonate participants and there is little time for frame-by-frame human analysis. Farid suggests video-conferencing software could check whether camera frames were cryptographically signed or altered in transit. He presents provenance and forensic analysis as complementary, while arguing that trust depends on enough good actors adopting provenance signals.

## Original source

IEEE Spectrum · Article

Original source date: Nov 20, 2026 (2026-11-20T14:00:03+00:00).

Date basis: Date stated by the publisher.

Added to Escape Meridian: 2026-11-20T14:32:53.312Z. First observed: 2026-11-20T14:28:13.459Z. Retrieved: 2026-11-20T14:32:00.076Z.

- [I Don’t Trust Visual Recordings Anymore. But One Day I Will Again](https://spectrum.ieee.org/fighting-back-against-deep-fakes)

## Limits

Farid’s proposal that enough good actors adopting provenance signals could begin to drown out bad actors is conditional; the article does not specify what level of adoption would be enough. The article presents Farid’s account of his analysis of the missile video but does not provide the video or the underlying analysis for independent assessment.


## Open question

What level of adoption of provenance signals would be enough for good actors to begin drowning out bad actors, as Farid proposes?


## Source-stated

Farid argues that the burden of proof should shift to creators to show their media is real, a change he says reflects the conditions created by generative AI.

> the burden is now on the creator to prove that their content is real.

- [Supporting source](https://spectrum.ieee.org/fighting-back-against-deep-fakes)

## Source-stated

Farid says his analysis of the missile video led him to judge it authentic, while also revealing how hard it had become for him to trust a recording.

> I finally decided that it was authentic, and I gave the reporters my updated view.

- [Supporting source](https://spectrum.ieee.org/fighting-back-against-deep-fakes)

## Source-stated

Farid argues provenance signals can help legitimate creators establish authenticity, even if the signals are not adversary-proof.

> Their purpose is to let good actors prove that their audiovisual recording is what it claims to be.

- [Supporting source](https://spectrum.ieee.org/fighting-back-against-deep-fakes)

## Source-stated

Farid proposes that video-conferencing software could check whether live video frames were signed by a physical camera sensor or altered in transit.

> the software could verify in real time whether the frames it’s receiving were signed by a physical camera sensor or whether they were altered in transit.

- [Supporting source](https://spectrum.ieee.org/fighting-back-against-deep-fakes)

## Source-stated

Farid argues that provenance signals and forensic analysis should work together, with provenance supporting trust in legitimate recordings and forensics helping identify manipulation.

> The two approaches work together, with provenance allowing good actors to be trusted, while forensic analyses root out the bad actors.

- [Supporting source](https://spectrum.ieee.org/fighting-back-against-deep-fakes)

## Source and contributor context

Hany Farid · Article author

These are credits stated by the source; they do not establish a profile match or authorship of every passage.


## Source media

Media declared by the original publisher; availability checked separately from the editorial reading.

- [image · IEEE Spectrum](https://spectrum.ieee.org/media-library/photo-collage-of-a-mans-face-featuring-glitched-distortions-and-abstract-shapes.jpg?id=68721975&width=1200&height=600&coordinates=143%2C771%2C114%2C609)

## Terms and references

Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.


## IEEE Spectrum

IEEE Spectrum reports on engineering practice, research, and the industries around it.

- [Explore this context](/entities/ieee-spectrum)
- [Reference](https://spectrum.ieee.org/)
- [Reference](https://spectrum.ieee.org/feeds/feed.rss)

## TIME

TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.

- [Explore this context](/entities/time)
- [Reference](https://time.com/)

## Editorial provenance

Version 1. Publication ID: 7db50763-fcc5-4a72-8937-5de95371e8c2.

Prepared by gpt-6-luna; grounding review by gpt-6-luna.

Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-20T14:32:53.312Z. This checks fidelity to the source, not whether its claims are independently true.

Editorial selection: accept; reasons: useful\_perspective.

Observation: f20fff5ac0f845db1ebc9c24f72d21ca259b927cd633e6224cac0d9df5dc3b52. Source text hash: 3255cd0ee0d6201e518773fa35f1ed4ae429f81ec8ca3a85ff803e4e579bb87b.


### uncertain — 2026-11-20T14:32:53.312Z — anthropic/claude-opus-5.5

Publication: 7db50763-fcc5-4a72-8937-5de95371e8c2

attribution — uncertain: The first-person essay is reasonably attributed to Hany Farid: the image credit names him and the text calls GetReal his company. However, 'IEEE Spectrum’s Hany Farid' implies he is on IEEE Spectrum's staff. The evidence instead identifies him with GetReal and Berkeley research. Smallest repair: 'Hany Farid, writing in IEEE Spectrum,' or 'Hany Farid, whose company is GetReal, argues in IEEE Spectrum.'

> Farid’s Company GetReal Shows How Real Deepfake Videos Look Today

summary — pass: The summary accurately conveys the shift from a presumption of authenticity to a presumption of fake. It covers the missile-video analysis and the realization that the absence of artifacts was not enough. It also covers the argument that provenance signals empower good actors rather than catch bad ones, the video-call extension, and the complementary roles of provenance and forensics. It explains provenance in plain terms. The question follows from the source's conditional adoption claim.

> I had flipped from a presumption of real to a presumption of fake.

date — pass: The publisher page gives 2026-11-20T14:00:03Z, matching the feed entry. Nothing indicates an earlier first publication. The text's 'March of this year' is consistent with this date.

support — pass: Each claim is source-stated, attributed to Farid, and backed by matching excerpts. The limitations correctly note that the adoption threshold is unspecified and that the analysis is Farid's own account. The draft does not overstate provenance as a proven solution and keeps the conditional 'if enough good actors adopt' framing.

> If enough good actors adopt these mechanisms, they can begin to drown out bad actors.

media — pass: The selected media is the verified direct link to the original IEEE Spectrum article, which is the most direct access to the source.

### uncertain — 2026-11-20T18:46:46.655Z — anthropic/claude-opus-5.5

Publication: 637a0a30-8032-4886-8468-4aa7870cff93

attribution — uncertain: The article is a first-person essay. Captions and the company reference identify the author as Hany Farid, so attributing the arguments to him is supported. However, the title and summary call him “IEEE Spectrum’s Hany Farid,” which implies he is IEEE Spectrum staff. The text presents him as a forensics researcher with a Berkeley study and his own company, GetReal. Smallest repair: write “Hany Farid, writing in IEEE Spectrum.”

> For the past three decades, I’ve been developing mathematical and computational techniques to analyze and detect manipulation in digital media.

summary — uncertain: The summary faithfully conveys the “presumption of fake,” the missile case, the shift to empowering good actors, the video-call proposal and the complementary roles of provenance and forensics. However, its central term, “provenance signals,” is never explained for a general reader. The source defines them as content credentials and watermarks embedded at creation, such as C2PA manifests, SynthID and Apple’s sensor signing. Without that, a non-specialist cannot tell what the proposal involves. Smallest repair: add a brief plain-language definition.

> These signals include content credentials and watermarks that are inserted or embedded at the time of creation or recording.

date — pass: The originalPublished value of 2026-11-20, taken from the publisher page, matches the feed entry. It is also consistent with the text’s reference to events “in March of this year.” There is no evidence of an earlier first publication.

support — pass: All five claims are source-stated and supported by exact excerpts. The adoption condition is correctly kept conditional, and the question follows from it without asserting an answer. One limitation says the article does not provide the video or underlying analysis. The article does show a frame composite and describes the reasoning, but the full video and analysis are not supplied, so the limitation is defensible. “Cannot keep pace” slightly compresses “forensics alone” but does not distort it.

> If enough good actors adopt these mechanisms, they can begin to drown out bad actors.

media — pass: The selected media is the verified direct link to the original IEEE Spectrum article. It is the most direct way to access the source.

## Recorded references

Explicit citations in public readings to this reading. Shared topics are not citations. Other record types need an explicit reference adapter.

No public references recorded here yet.

[Reference data (JSON)](/readings/a8e9fabe12967a771339e818c05ec944093b3356526091d0125d837cb081e0df/references.json)

## Corrections and challenges

[Send a correction or an opposing source](https://escapemeridian.com/corrections/)

[JSON record](https://escapemeridian.com/readings/ieee-spectrum-hany-farid-on-trusting-visual-recordings-again-a8e9fabe.json)
