SCANT: A (kind-of-decent) Framework for Ethical Deepfake Creation & Distribution

Contents

1. The Ethical Blueprint: Building Trust in Synthetic Media
   1. S - Social Benefit
   2. C - Consent
   3. A - Accountability
   4. N - Non-Deception
   5. T - Transparency
2. Putting SCANT into Practice
3. TL;DR Checklist
4. It takes work!
5. AI - Embracing the Human
6. Speaking of ISO 42001

The Ethical Blueprint: Building Trust in Synthetic Media

Lots of damage has been done with AI, and to keep from deep-sixing the forward-leaning tone I want in this article, I’ll refrain from noting any details – the internet is available for you to search to your heart’s content. I want to start with that note because how we use AI is not just an option, like whether we want a cinnamon roll or a bagel at breakfast. AI use has meaning – whether it’s dark or not depends on each of us.

On the lighter side of negative consequences, more and more media influencers are posting AI videos claiming that those videos aren’t AI, just to increase their views and to increase interaction – drawing out those who claim, rightly, that the media truly is AI, and then arguing back and forth about its validity. Some influencers are abusing AI to waste peoples’ time for the sole benefit of the influencer; and those actions a) further erode viewers’ trust in media platforms and b) turn them against the hope of the real usefulness of AI. (plus, it wastes their time, and that time is part of life, so it really grinds on the nerves to realize that one has spent some of their life’s breath only to be been taken for a fool). So, even on the lighter side, those consequences are eroding trust in all-things-online. But with running with the concept of “learn to discern,” one can beat that fraud. Mitch Clark does an excellent job at educating on this.

For a recent cybersecurity presentation, I worked with the bots to create an ethical AI framework to fit GenAI in general. The bots gave me some decent ideas, and I prompted back and forth with them, and then moved it into my own short set that I then made into an acronym. Mnemonic devices to the rescue! This is also an example of human-in-the-loop in AI – the robot gives some information, and there’s back-and-forth between me and the machines, but in the end it’s human creativity and alignment of the information presented that wins the day.

Are there other frameworks? Sure! Well-researched ones, official governmental ones, professional community-developed ones. But I wanted to make one that’s maybe more accessible to the general public. Will it fail to gain traction? Certainly! But I have today, and maybe someone will read this and either learn something, or think “I can do better” and I will have accomplished a goal of forcing others to write better things because I wrote a so-so thing. (I do this at home – I throw out weird, and even bad ideas, and that forces the kids to create better ideas 😊 )

SCANT is a concise and actionable set of principles designed to keep AI-generated media safe, respectful, and trustworthy.

Why SCANT? Scant means “falling short of what is normal, necessary, or desirable.” I decided to keep it because it’s an insufficient approach, but may be simple enough to either work as-is or urge others on to make their own. AI is still a nascent field, and a good improvement would be for those involved in the field to either adopt or form their own workable models for evaluating how AI is developed in their org, even if it’s nothing fancy.

Yes, of course there’s ISO 42001 ! There are great things happening around the world. Those can be expensive and cumbersome, though just knowing the principles and proceeding accordingly is a great way to show others how you align your AI practices with the international standard. SCANT is simply a “pet project,” if you will, so I thought I’d bring it in the open.

NOTE: The term “deepfake” is often used to describe unethical use of high-quality GenAI, but the term is actually used for any of those high quality results. Throughout this article, deepfake is used for the final product, not simply for the deceptive kind.

SCANT

S - Social Benefit

The Goal? Deploy deepfake technology only when it creates a positive impact for individuals, communities, or society at large.

C - Consent

Goal: Secure explicit and informed permission from every person whose likeness, voice, or mannerisms are used.
(Are voice trademarks the path forward? See this article re: Matthew McConaughey trademarking his voice https://analystip.com/matthew-mcconaughey-trademark-himself-to-stop-ai-clones/ )

A - Accountability

Goal: Ensure that creators, distributors, and platform operators can be identified, held responsible, and answerable for the consequences of synthetic media.

N - Non‑Deception

Goal: Prevent the intentional misleading of viewers; synthetic media should be used to educate, entertain, or augment reality - not to fabricate false narratives.

T - Transparency

Goal: Make it unmistakably clear that a piece of media is synthetically generated; and disclose the technical provenance behind it.

Putting SCANT into Practice - a mini-workflow

1. Idea Generation
   • Conduct a Social Benefit questionnaire
   • Draft a Purpose Statement and identify target audiences
2. Consent Acquisition
   • Send a consent package (explanation + digital signature)
   • Store signed consent with timestamped metadata
3. Model Selection & Documentation
   • Choose an AI model, note version, training data, and any fine‑tuning.
   • Log all generation parameters in a tamper‑evident ledger
4. Creation & Attribution
   • Generate the deepfake, embed provenance metadata, and apply a visual Transparency watermark
   • Record the creator’s identity and any human post‑processing
5. Review & Accountability Check
   • Submit the asset to the internal Ethics Board for Non‑Deception and Social Benefit validation
   • Verify that Consent and Transparency requirements are satisfied
6. Publication
   • Release the content on chosen platforms with clear labeling (e.g., “Synthetic Media – For Educational Purposes”)
   • Provide a public link to the provenance record
7. Post‑Release Monitoring
   • Track audience reactions, complaints, or misuse reports
   • If a violation is detected, invoke the Accountability remediation plan (removal, apology, compensation)

TL;DR Checklist

• Social Benefit: Purpose‑first, impact‑focused, measurable outcomes.
• Consent: Informed, freely given, specific, revocable, verified.
• Accountability: Provenance metadata, governance board, liability clauses, audit logs, remediation pathways.
• Non‑Deception: Intent clarity, no covert splicing, preserve factual context, educate viewers.
• Transparency: Persistent synthetic label, model/version disclosure, generation parameters, human‑oversight note, public provenance API, visual watermark.

Embedding these SCANT principles into every stage of the GenAI lifecycle - from conception to post‑release monitoring - creates a defensible (yes - you'll need to defend it to customers and/or auditors) and ethically sound workflow that respects individuals, protects public discourse, and unlocks the creative and societal potential of synthetic media.

How does the typical person use this? It’s really easy, actually (at least, it should be). Don’t be a jerk, don’t steal, don’t defraud, don’t mislead. Those bad things have been going on for a long time. Anyone on any social media has seen or taken part in news by memes, spreading mis/dis/mal information simply by passing on something revealing or shocking based only on a well-crafted meme, which probably has authority based on who shared it, but itself has no link, source, time/date stamp, or anything that remotely resembles a way to verify. Don't be part of that "game" - help out viewers and readers.

Humor, comedy, and satire are part of freedom of speech, so it's vital that we remain ethical in synthetic media while NOT stamping out these freedoms.

It takes work!

You may notice that this is a lot of work. It should be. The capabilities of AI can’t be taken lightly. Lives and reputations have been harmed because of the focus on Speed over Stability. While much of the software world has been driven by, “Hey, let’s create something simple and then build on it after we get feedback,” so much of AI has been, “Let’s throw this enormous thing out that, see what happens, and then pare back from there.”

Whatever approach is taken, the work needs to be put in to make AI a proper tool for everyone who wants to use it.

AI - Embracing the Human

AI is powerful. Proceed cautiously. Any reputable AI offering will use cautionary words similar to, “Do not rely on AI’s results for critical business or personal decisions.” Everyone knows that AI can be wrong…or do they? (think of the movie, "I, Robot")

AI does not have intellect or will – it has no soul. It’s complicated, hi-tech, seemingly magical and mystical, but behind it all is a robot, a bunch of machines.

Harvard has an AI for Human Flourishing program, so that’s a good way to get insight into in-depth studies already performed on the potential full useful aim of AI.

An AI benchmark in conjunction with this program is the Flourishing AI Benchmark (FAI Benchmark). The benchmark is described as follows:

“…a novel benchmarking approach that evaluates LLMs across seven key dimensions of human flourishing, based on the flourishing measure developed by researchers at the Human Flourishing Program at Harvard and in collaboration with Barna and Gloo:

1. Character and Virtue (Character)

2. Close Social Relationships (Relationships)

3. Happiness and Life Satisfaction (Happiness)

4. Meaning and Purpose (Meaning)

5. Mental and Physical Health (Health)

6. Financial and Material Stability (Finances)

7. Faith and Spirituality (Faith)”

A name connected to the study is Pat Gelsinger, a name you may recognize from his times as CEO of VMWare and Intel. More information here: https://techcrunch.com/2025/07/10/former-intel-ceo-launches-a-benchmark-to-measure-ai-alignment/

From their report: “Initial testing of 28 leading language models reveals that while some models approach holistic alignment (with the highest-scoring models achieving 72/100), none are acceptably aligned across all dimensions, particularly in Faith and Spirituality, Character and Virtue, and Meaning and Purpose.”

Don’t expect AI to speak realistically to matters pertaining to the above deficient areas. Regrettably, many have been severely and negatively affected by expecting their LLM(s) to provide precisely these kinds of ethical and life guidance.

In general, current models have the goal of helping people while not causing harm, but some important areas have been overlooked. With such powerful implications, while AI has not considered people holistically, it could be developed to take the whole person into account.

 Speaking of ISO 42001

ISO 42001 is considered THE international standard and governance framework for AI. To align with ISO 42001 principles, here’s a high-level plausible mapping (focus on plausible - nothing official here).