Please ensure Javascript is enabled for purposes of website accessibility
Back

White House Secures Voluntary Pledges from AI Firms on Deepfake Safeguards

Doron Ish Shalom
,
Head of BizDev & Strategic Partnerships
September 14, 2024 • 2 minutes read

The rise of deepfakes, particularly those exploiting individuals for non-consensual pornography and spreading malicious misinformation, has become a pressing concern for policymakers and the public alike.

Recognizing the urgent need for action, the Biden-Harris administration has successfully secured voluntary commitments from leading artificial intelligence companies to implement safeguards against the creation and dissemination of harmful deepfakes.

This initiative marks a significant step towards mitigating the potential societal damage posed by unchecked AI-generated content.

 

White House Agreement

The White House's effort focuses on obtaining voluntary pledges from key players in the AI industry, including OpenAI, Anthropic, and Microsoft.

These commitments aim to address aspects of deepfake misuse: the creation of non-consensual intimate imagery, often referred to as deepfake nudes, and the spread of AI-generated misinformation.

The administration, led by figures like Arati Prabhakar, Director of the White House Office of Science and Technology Policy, has emphasized the urgency of these measures, highlighting the potential for deepfakes to erode trust in media, manipulate public opinion, and inflict emotional harm on victims.

 

The Core Commitments: Technical and Policy Safeguards 

Voluntary commitments include a range of technical and policy-based safeguards. On the technical side, companies have pledged to develop and implement systems for watermarking or labeling AI-generated content, making it easier to identify manipulated media. They are also working to enhance detection algorithms capable of identifying deepfakes with greater accuracy. Crucially, these firms are taking measures to prevent the generation of deepfake sexual content by implementing filters and restrictions within their AI models.

Beyond technical measures, the commitments extend to policy and enforcement. Companies have agreed to strengthen their terms of service to explicitly prohibit the creation and distribution of deepfakes for abusive purposes.

They are also exploring avenues for collaboration with law enforcement and other agencies to address deepfake-related crimes. Furthermore, transparency and accountability are key components of the initiative.

Societal Impact and the Future of AI Ethics

The implications of these voluntary pledges extend beyond the immediate goal of curbing deepfake abuse. They have the potential to shape the future of AI development, striking a balance between innovation and ethical considerations.

By proactively addressing the risks associated with AI-generated content, these companies can foster greater public trust in the technology. However, the effectiveness of voluntary commitments remains controversial.

While they demonstrate a willingness on the part of industry leaders to address the issue, some argue that stronger regulatory measures may be necessary to ensure consistent and enforceable standards.

“Seeing big tech voluntarily agree on AI safety is encouraging – especially the pledge to watermark AI content. It’s a sign that both industry and government recognize the deepfake threat and are beginning to grow in the same direction to address it.”

Latest AI Deepfake articles

Deepfake Investment Scams Are Exploding—And the Stakes Just Got Personal

Over the past few weeks, my feed has been flooded with "exclusive" video pitches featuring familiar faces like Gal Gadot, Dovi Frances, Yasmin Lukatz, Eyal Valdman, and even Warren Buffett. Each video promises extraordinary returns from a supposedly exclusive investment fund. The presentations are incredibly polished, flawlessly lip-synced, and convincingly authentic.

The only problem? None of these videos are real.

Why Does This Matter?

  • Hyper-Realism on Demand: Advanced generative AI now easily replicates faces, voices, and micro-expressions in real-time.
  • Massive Reach: Fraudsters distribute thousands of micro-targeted ads across Instagram, YouTube Shorts, and TikTok. Removing one only leads to a rapid replacement.
  • Record Losses: In 2024, a deepfake impersonation of a CFO cost a UK engineering firm $25 million. Regulators estimate nearly 40% of last year's investment fraud complaints involved manipulated audio or video.

What To Watch For

  • Too-Good-To-Be-True Promises: Genuine celebrities rarely endorse 15% daily returns.
  • One-Way Communication: Disabled comments, invitation-only direct messages, and suspiciously new "official" websites are red flags.
  • Subtle Visual Artifacts: Watch for flat hairline lighting, inconsistent blinking patterns, or an unnatural stare when the speaker moves.

How Clarity Responds

At Clarity, our detection engine swiftly identified the recent "Gal Gadot investment pitch" deepfake within 4 seconds, pinpointing subtle lip-sync inconsistencies invisible to human observers.

As deepfakes proliferate at machine speed, automated verification is essential. Our technology analyzes facial dynamics, audio patterns, and metadata in real-time, enabling rapid removal of fraudulent content—before it reaches potential victims. Think of our solution as antivirus software for the age of synthetic media—always active, continuously evolving, and most effective when supported by an educated public.

Yet, technology alone isn't enough; critical thinking and vigilance remain crucial.

If You Encounter a Suspicious Investment Video:

  • Pause: Don’t act immediately.
  • Verify: Confirm the source through known, official channels.
  • Report: Use the “impersonation” option available on most platforms.
  • Share Awareness: Inform others. Community awareness grows faster than deepfake scams when actively spread.
Together, let's protect our communities—investors, families, and fans alike—from synthetic media fraud.
graphical user interface, website

Last week, Unit42 by Palo Alto Networks published a fascinating - and frightening - deep dive into how easily threat actors are creating synthetic identities to infiltrate organizations.

We’re talking about AI-generated personas, complete with fake resumes, social profiles, and most notably, deepfaked video interviews. These attackers aren’t just sending phishing emails anymore. They’re showing up on your video calls, looking and sounding like the perfect candidate.

At Clarity, this is exactly the kind of threat we’ve been preparing for.

The Rise of Deepfakes in Hiring - A New Attack Vector

The interview process has become a weak link in organizational security. With remote hiring now standard, verifying a candidate’s identity has never been more challenging - and adversaries know it.

Deepfake technology has reached a point where bad actors can spin up convincing video personas in hours. As Unit42 highlighted, state-sponsored groups are already exploiting this to gain insider access to critical infrastructure, data, and intellectual property.

This isn’t just a cybersecurity issue - it’s a trust crisis.

Inside Unit42’s Findings - A Manual Deepfake Hunt

In their detailed analysis, Unit42 showcased just how layered and complex synthetic identity attacks can be. Each figure in their report highlights different aspects of deepfake deception - from AI-generated profile photos and fabricated resumes to manipulated video interviews, with cheap and widely available hardware to higher-quality deepfakes using resource-intensive techniques.

Their approach demonstrates the painstaking process of manually dissecting these fakes:

  • Spotting subtle visual glitches

  • Identifying inconsistencies across frames

  • Cross-referencing digital footprints

While their expertise is impressive, it also underscores a critical point: most organizations don’t have the time, resources, or deepfake specialists to conduct this level of forensic analysis for every candidate or call.

That’s exactly why Clarity exists.

How Clarity Detects What the Human Eye Can’t

Let’s face it - no recruiter, hiring manager, or IT admin can be expected to spot a high-quality deepfake in a live interview. That’s where Clarity comes in.

Our AI-powered detection platform is designed to seamlessly analyze video feeds, pre-recorded interviews, and live calls to identify synthetic media in real-time.

When we ran the videos shared in Unit42’s report through our Clarity Studio, the outcome was clear:

Deepfake detected - with a clear confidence score that tells you instantly whether a video is real or synthetic. No need for manual checks or deepfake expertise - Clarity delivers fast, decisive answers when it matters most.

No manual frame-by-frame reviews. No specialized training required. Just fast, reliable detection that integrates directly into your workflows.

Automating Trust in a Synthetic World

At Clarity, we believe organizations shouldn’t have to become deepfake experts to stay protected. Whether you're hiring globally, conducting sensitive interviews, or verifying identities remotely, our system ensures:

  • Real-time detection during live calls

  • Comprehensive analysis of recorded videos

  • Automated alerts when synthetic media is detected

With Clarity, you can focus on growing your team and business, without second-guessing who’s really on the other side of the screen.

See It In Action

We applaud Unit42 for shedding light on this growing threat. To demonstrate how proactive detection can neutralize these risks, we’ve analyzed the same deepfake videos from their post using Clarity Studio.

Check out the screenshots below to see how Clarity instantly flags these synthetic identities - before they become your next insider threat.

Our studio results on Unit42 Figure 4 video: A demonstration of a realtime deepfake on cheap and widely-available hardware

Our studio results on Unit42 Figure 4 video: A demonstration of a realtime deepfake on cheap and widely-available hardware
Our studio results on Unit42 Figure 5: demonstration of identity switching
Our studio results on Unit42 Figure 6. A higher quality deepfake using a more resource-intensive technique
Our studio results on Unit42 Figure 7c. The "sky-or-ground"

On Saturday night, Israeli Channel 14 mistakenly aired a manipulated video of former Defense Minister Yoav Gallant—an AI-generated deepfake that appeared to originate from Iranian media sources. The incident, which took place during the channel’s evening newscast, showcased Gallant speaking in Hebrew but with a clear Persian accent. The anchor, recognizing the suspicious nature of the clip, interrupted the broadcast mid-sentence, calling out the video as fabricated.

“On the first sentence I said stop the video. We apologize. This is cooked… These are not Gallant’s words but AI trying to insert messages about the U.S. and the Houthis,” said anchor Sarah Beck live on air.

Shortly after, Channel 14 issued an official statement confirming that the video was aired without prior verification and that an internal investigation was underway.

What Actually Happened?

The video portrayed Gallant stating that “the U.S. will not be able to defeat the Houthis,” a politically charged statement intended to sow confusion and manipulate public sentiment. Although the channel removed the clip within seconds, the damage was already done: the AI-generated video had reached thousands of viewers.

This incident highlights the speed, sophistication, and geopolitical implications of deepfake attacks.

How Clarity Responded — in Real Time

Minutes after the clip aired, our team at Clarity ran the footage through Clarity Studio, our real-time media analysis and deepfake detection platform. The results were clear:

  • Manipulation Level: High
  • Audio-Visual Inconsistencies: Detected in voice pattern and facial dynamics
  • Anomaly Source: Synthetic voice generation with foreign accent simulation

Here’s the detection screenshot from Clarity Studio:

We identified clear mismatches between Gallant’s known voice and speech pattern compared to the clip, along with temporal inconsistencies in facial movement and audio syncing—hallmarks of state-sponsored deepfake manipulation.

Why It Matters

This wasn’t a fringe incident. This was a high-profile deception attempt broadcast on national television. Deepfakes are no longer future threats. They are present-day weapons—used to spread disinformation, manipulate public opinion, and erode trust in media.

And this time, Clarity caught it before the narrative could spiral out of control.

The Takeaway

Broadcasters, law enforcement, and government agencies need tools that can verify audio and video authenticity in real time. This isn’t just about technology—it’s about safeguarding democratic discourse and preventing psychological operations from hostile actors.

At Clarity, we’re building the tools to detect these threats before they become headlines.