<p>Meta has unveiled Movie Gen, a suite of AI models capable of generating and editing video, audio, and images, including creating realistic videos from a single photo. While the models reportedly outperform competitors in human evaluations, concerns about misuse, particularly in creating deepfakes, persist. Movie Gen allows personalized video creation and sound synthesis, enabling detailed edits and soundscapes. Meta aims to enhance creativity without replacing artists, but faces challenges regarding the ethical implications of its technology</p>
<p>Read the <a href="https://arstechnica.com/ai/2024/10/metas-new-movie-gen-ai-system-can-deepfake-video-from-a-single-photo/" rel="noopener noreferrer" target="_blank">full article</a> to learn more.</p>

Meta’s new “Movie Gen” AI system can deepfake video from a single photo

<p>Deepfake incidents are expected to rise by 60% in 2024, with over 150,000 cases globally. Deloitte projects that these attacks will cause $40 billion in damages by 2027, primarily targeting banking and finance. Nation-states are increasingly using AI-generated deepfakes in cyberwarfare, causing concerns about trust in institutions and elections. Models like OpenAI’s GPT-4o are designed to detect deepfakes using advanced techniques like Generative Adversarial Networks (GANs), voice authentication, and multimodal cross-validation to counter the growing threat of AI-powered deception.</p>
<p>Read the <a href="https://venturebeat.com/security/how-gpt-4o-defends-your-identity-against-ai-generated-deepfakes/" rel="noopener noreferrer" target="_blank">full article</a> to learn more.</p>

GPT-4o: OpenAI’s shield against $40B deepfake threat to enterprises

<p>U.S. Senator Ben Cardin was recently targeted by a deepfake operation impersonating Ukraine’s former Foreign Minister, Dmytro Kuleba. The deepfake, which sought to manipulate the senator into discussing sensitive political topics related to the upcoming U.S. Presidential election, appeared highly convincing. Cardin, suspicious of the interaction, cut the call and reported the incident, which is now under FBI investigation. This event follows previous cases of deepfakes used in attempts to deceive Western political figures, underscoring the growing threat of AI-generated impersonations in both politics and business.</p>
<p>Read the <a href="https://www.bitdefender.com/blog/hotforsecurity/deepfake-ukrainian-diplomat-targeted-us-senator-on-zoom-call/" rel="noopener noreferrer" target="_blank">full article</a> to learn more.</p>

Deepfake Ukrainian diplomat targeted US senator on Zoom call

<p>When Deep-Live-Cam first launched in April of 2024, it created immediate shockwaves: Only weeks earlier, Microsoft teased its VASA-1 research that creates realistic gen-AI generated videos from just a single image, with no immediate intention to productize it.</p>
<h2>What is Deep-Live-Cam?</h2>
<p><strong>Deep-Live-Cam</strong> is an open-source tool, available for free on GitHub, that allows anyone to create real-time face swaps of high quality, look 3D and lifelike, and with only a single source image required - available right now and free of charge. Designed as a creativity tool to empower artists with a simplified animation process, crafting unique avatars and improving productivity for creative professionals.</p>
<p>But just like you, who reads this blog on cybersecurity, our first thought was that Deep-Live-Cam has ushered in a new era of AI-generated content and, with that, the obvious and truly significant challenge of deepfake detection: With even <a href="/blog/open-ai-quietly-shuts-down-its-ai-detection-tool">OpenAI shutting down its own Gen-AI detection tools in the past due to disappointing ability to detect its own creations</a>, it is the current state of technology that even the leaders of GenAI technology struggle to reliably identify its own outputs - which makes us wonder where that leaves companies that offer Deepfake Detection tools, let alone their enterprise customers who need to be able to detect Gen-AI content to protect their customers from social engineering and identity theft.</p>
<p>For this blog, we created a video using the Deep-Live-Cam library and tested it with the popular “<a href="https://scanner.deepware.ai/" rel="noopener noreferrer" target="_blank">Deepware Deepfake Video Scanner</a>,” which is the top Google result for deepfake scanners or detection tools as of December 2024.</p>
<h2>How did detection algorithms perform against Deep-Live-Cam?</h2>
<p>First, I uploaded a video generated with Deep-Live-Cam. Deepware Scanner then uses four different detection algorithms, each of which outputs a probabilistic score of how likely the uploaded video is to contain a deepfake.</p>
<p>Here’s how the detection tools fared:</p>
<h3>1. Avatarify Detection</h3>
<ul>
<li>Result: <strong>NO DEEPFAKE DETECTED (1%)</strong></li>
<li>In production, this algorithm would result in 99% false positives.</li>
<li>The Avatarify algorithm showed a negligible 1% chance that the video was a deepfake, nearly certain it was authentic.</li>
</ul>
<h3>2. Deepware Detection</h3>
<ul>
<li>Result: <strong>NO DEEPFAKE DETECTED (3%)</strong></li>
<li>In production, this algorithm would result in 97% false positives.</li>
<li>Deepware flagged only a 3% probability of manipulation, indicating high confidence in the video’s authenticity.</li>
</ul>
<h3>3. Seferbekov Detection</h3>
<ul>
<li>Result: <strong>NO DEEPFAKE DETECTED (28%)</strong></li>
<li>In production, this algorithm would result in 72% false positives.</li>
<li>This model performed the best among the four. However, it still has a significant failure rate that would be unacceptable if used to detect fraud.</li>
</ul>
<h3>4. Ensemble Detection</h3>
<ul>
<li>Result: <strong>NO DEEPFAKE DETECTED (9%)</strong></li>
<li>In production, this algorithm would result in 91% false positives</li>
<li>The Ensemble model, which combines multiple detection approaches, detected only a 9% likelihood of deepfake content.</li>
</ul>
<p>These results reveal that Deep-Live-Cam can create videos so convincing that existing algorithms fail to classify them accurately, even when specifically designed to identify fake media.</p>
<h2>Why Is Deep-Live-Cam So Convincing?</h2>
<h3>1. Real-Time Performance</h3>
<p>Deep-Live-Cam creates deepfakes at high framerates in real time to be shared like a simple Snapchat filter, except that it creates complex, three-dimensional impressions of the impersonated avatar.</p>
<h3>2. Micro-Expression Precision</h3>
<p>The technology captures and reproduces subtle expressions like blinking, lip movements, and lighting variations with uncanny accuracy. It employs neural networks that refine artifacts in real-time, eliminating common cues such as unnatural movements or lighting mismatches.</p>
<h3>3. Stream-Friendly Outputs</h3>
<p>Videos streamed or compressed for live settings add noise that actively masks telltale signs. Accurate deepfake detection highly depends on the video quality, but there are many legitimate reasons why real video streams may have noise or compressed quality,  further complicating detection.</p>
<h2>The Implications of Undetectable Deepfakes</h2>
<h3>1. A Looming Trust Crisis</h3>
<p>Videos have long been considered credible evidence, but tools like Deep-Live-Cam undermine this trust rapidly. If a court case is presented with crucial evidence in the form of recorded video, but that video could have been manipulated by an intentional impersonator, how would we decide whether to dismiss it as evidence when detection tools fail to keep pace?</p>
<h3>2. Existing Fraud Risks are Amplified</h3>
<p>Undetectable deepfakes already have and will increasingly enable fraud, impersonation, and misinformation campaigns, posing significant risks to cybersecurity and public trust. We will face the same trust challenges as before but amplified by fraudsters’ ability to convincingly impersonate anyone and a lack of tools to accurately scan for such fraud attacks.</p>
<h3>3. Regulation will be Reactive</h3>
<p>Existing laws and policies struggle to address the ethical and legal implications of real-time, undetectable deepfakes. An example of this is that <a href="https://www.usaemploymentlawyers.com/blog/2023/february/differences-between-pii-and-biometric-data/" rel="noopener noreferrer" target="_blank">“biometrics” are not considered PII</a> (Personally identifiable information), unlike e.g., your social security number. While it sounds surprising in the context of deepfakes, the reason for biometrics (like your facial likeness) not being PII was that you can claim your name or SSN is that of another person, but it’s impossible to claim your face looks like the face of the person next to you - this reality has drastically changed in the context of real-time deepfakes and remote work.</p>
<h2>What’s Next?</h2>
<p>As detection of technology like Deep-Live-Cam falls short, companies need to adopt a multi-pronged approach to ensure security while regulations and technology catch up:</p>
<h3>Protect the Root:</h3>
<ul>
<li>Safas Secure Camera to prevent content injection: We wouldn’t write a blog if we didn’t have something relevant to contribute. Our Secure Camera’s purpose is to certify the authenticity of the footage as it exits the camera all the way to the application in which it is displayed, effectively removing any open door for inserting deepfakes or virtual cameras into the loop, for impersonation or other reasons.</li>
</ul>
<h3>Authenticate the material:</h3>
<ul>
<li>Biometric authentication of meeting participants: Embed biometric verification protocols before admitting users, for example, by using authenticators or face biometric authenticators that need to be passed before a participant can join confidential meetings. Biometric authentication relies on probabilistic scanning, so there is the risk of false positives (undetected deepfakes). While we can expect deepfake detection models to improve, it is a race of AI models against GenAI models, and it is recommended not to rely on it exclusively.</li>
</ul>
<h3>Raise awareness internally</h3>
<ul>
<li>Any CISO and cybersecurity advocate knows this: Most fraud succeeds when the victim is not questioning an unusual request or is unaware of the red flags that should make them second-guess an action. Education about what is possible is crucial. One approach can be to share blogs like this one to educate your staff on how convincing deepfakes already look and how cheap and easy they are to create. Ensure that no staff member believes deepfakes are either exceptionally rare and they’d probably recognize one right away.</li>
</ul>
<h2>Conclusion</h2>
<p>Deep-Live-Cam showcases the immense creative potential of AI, and with that, also provides an eery preview of the tools that are available to fraudsters today in creating highly realistic synthetic media.</p>
<p>This blog demonstrated how the inability of tools like Avatarify, Deepware, Seferbekov, and Ensemble to detect Deep-Live-Cam presents a dangerous challenge for society. As deepfake technology becomes increasingly sophisticated, so must our efforts to develop detection, authentication, and regulatory solutions.</p>
<p>Safas is currently in closed beta.  <a href="/waitlist">Join our waitlist</a> for a trial.</p>

Deep-Live-Cam: High-quality deepfake avatars, free for anyone

<p>Amidst controversial return-to-office mandates by large companies like Amazon, remote and hybrid work is actually still growing - remote work is up 6% in the US since 2023, according to Owl Labs’s 2024 “State of Hybrid Work” report (<a href="https://owllabs.com/state-of-hybrid-work/2024" rel="noopener noreferrer" target="_blank">source</a>).</p>
<p>And with remote work, video conferencing has solidified its permanence in handling every workplace interaction. - from daily standups to client meetings, hiring interviews and high-stakes financial transactions, virtual meetings on camera are here to stay. Yet, as companies open up to rely on platforms like Google Meets, Teams, Zoom, and WebEx, fraudsters focus increasing efforts on exploiting these new entry points in a new wave of cybersecurity threats.</p>
<h4>The stakes are high:</h4>
<p>Video conferencing, for all its convenience, has proven to be particularly vulnerable to sophisticated forms of digital attacks. With sophisticated deepfake technology already available as end-user services at low cost, attackers find new ways to manipulate what we see and hear on the screen. Deloitte recently published a study on deepfake fraud in financial services, highlighting how these threats are no longer just speculation of what’s to come - they’re real, they’re growing, and they’re affecting industries ranging from banking to education. Below, we break down the five most pressing threats to watch closely in 2025—and what security leaders can do to prepare.</p>
<h2>1. Injection of Pre-Recorded Video</h2>
<p><img src="/images/posts/top-5/recorder.jpg" alt="pre-recorded video injection"></p>
<p>One of the more traditional threat types involves attackers injecting pre-recorded video to make it appear as if a person is present and engaging in real-time. Prerecorded footage can also be manipulated to create infinite loops or otherwise be tampered with to go undetected.</p>
<h3>Why It’s a Threat:</h3>
<p>This tactic could let unauthorized users join a leaked meeting link and pose as legitimate attendees, violating security protocols and gaining access to sensitive discussions while the fraudster is listening. Imagine an internal company meeting in which an employee’s video is used to gain confidential insights about strategic or financial decisions.</p>
<h3>Prevention Strategy:</h3>
<p>Real-time hardware verification technology offers a robust first line of defense - coincidentally, this is what Safas does: It guarantees that camera feeds in video calls originate from real camera hardware, and are not the result of a video streamed via a virtual camera.</p>
<h2>2. Injection of Deepfakes (Face Swaps, Face Morphs, Synthetic faces)</h2>
<p>Deepfakes are imagery created using generative AI, allowing bad actors to alter or replace or morph their face with that of someone else: They are no longer a preview of what’s to come, but already here: free tools like Deep-Live-Cam are well-known in the online community on Reddit, and despite being free they work very effectively with only a single photo of the victim needed to assume their likeness. As a result, the injection of deepfakes has emerged as a top security threat.</p>
<h3>Why It’s a Threat:</h3>
<p>In a high-stakes environment—say, an M&#x26;A negotiation or a legal deposition—the consequences of a fraudster duping employees into transferring payments or leaking information have already had disastrous consequences and constitute regulatory compliance violations, financial and reputational loss in a few high profile cases, see our <a href="/blog/real-time-ai-generated-deepfakes-in-meetings">first blog</a>.</p>
<h3>Prevention Strategy:</h3>
<p>IT administrators can employ deepfake detection apps that are available on marketplaces like the Zoom App Marketplace. These apps scan video for deepfakes, which unfortunately is a race against rapidly evolving GenAI itself, and it can be easily spoofed by a fraudster lowering camera resolution (and blaming their Internet connection for it), among other tricks.
A more secure way is once again to guarantee the camera stream comes from the real camera. This is particularly crucial for meetings that can be accessed via web browsers, which are completely unprotected against video injection.</p>
<h2>3. Adding Filters to Hide The Real Image</h2>
<p>While the concept of filters often brings Snapchat or Instagram to mind, they’re becoming a security concern for video conferencing. Attackers use overlays or effects to obscure their real identity, disguising features or making it challenging to determine who they are. These filters are not necessarily created by GenAI, but they can be to make the effect more subtle when facing an unknown person (for example applying gender-change filters).</p>
<h3>Why It’s a Threat:</h3>
<p>Filters make it harder to verify someone’s true looks, enabling bad actors to sidestep doubts of the victim. This could become particularly problematic in industries where visual ID checks are part of the security process.</p>
<h3>Prevention Strategy:</h3>
<p>In this case, IT managers can enforce policies to only allow filters in video conferencing tools that do not hide the face of the person. Filters are overlays by apps such as Teams or Zoom, and can be configured, flagged and even be deactivated at the IT admin level. For filters that are already part of the camera feed, they may be maliciously added via a virtual camera.</p>
<h2>4. AI Voice Generation (Audio Deepfake)</h2>
<p>Deepfake technology doesn’t stop at video; audio deepfakes are on the rise, too, allowing for fraud without taking the risk of adding a visual. With the help of voice AI, attackers can mimic a person’s voice in real-time, opening the door to impersonation and manipulation through verbal exchanges.</p>
<h3>Why It’s a Threat:</h3>
<p>A skilled attacker could engage in real-time dialogue as someone else, potentially extracting confidential information or issuing harmful directives. Recent cases, such as one in which a PhD student spoofed a popular voice authentication system (<a href="https://www.biometricupdate.com/202307/phd-student-uses-deepfake-to-pass-popular-voice-authentication-and-spoof-detection-system" rel="noopener noreferrer" target="_blank">source</a>), highlight that this threat has materialized already.</p>
<h3>Prevention Strategy:</h3>
<p>Verifying voices against source samples can help counter this risk - though the risk is the same as with AI-driven deepfake detection: AI is rapidly improving, so any detection method needs to keep up with a rapidly evolving opponent that is headed for perfection. While it would be intuitive to say that Safas can help the real audio feed from the microphone be transmitted as part of a video call, another immediate remedy is to make video turned on a mandatory requirement for high-stakes conversations.</p>
<h2>5. Hijacking Meeting Links via Compromised Accounts</h2>
<p><img src="/images/posts/top-5/woman-typing-on-keyboard.jpg" alt="hijacking meeting links">
Finally, there’s the fundamental threat of meeting link hijacking, where attackers gain access to a participant’s account and join meetings as legitimate participants. Once in, they can use any of the tactics above to manipulate, deceive, or extract information.</p>
<h3>Why It’s a Threat:</h3>
<p>Hijacking access is a hacking approach as old as hacking itself. In video conferencing, this threat is taken to another level, as meeting links can be reverse-engineered (e.g. custom links that use employee names in certain formatting), and especially larger meetings can be attended by unauthorized parties. In sensitive environments, like corporate boardrooms or government briefings, the potential for damage is immense, as demonstrated by Russian spies who attended WebEx meetings with the German military that discussed their Ukraine strategy (<a href="https://www.theregister.com/2024/03/04/germany_confirms_russia_leak_genuine/" rel="noopener noreferrer" target="_blank">source</a>).</p>
<h3>Prevention Strategy:</h3>
<p>This is a fundamental security loophole, created by the rapid change to remote work. Organizations often missed to protect videoconferencing access with authentication to ensure end-to-end verification of all participants. Security leaders should consider using a secure camera feed using a solution like Safas, requiring MFA-login of all participants for high-risk meetings, or even procuring a biometric authenticator to join confidential meetings.</p>
<h2>Conclusion: It’s hard to win the race against AI - consider guardrails instead</h2>
<p>Video conferencing has become essential for remote work, bringing new security threats that many CISOs and compliance officers may overlook. With deepfakes now free and easy to make, guardrails are crucial to prevent these threats from becoming as common as spam mails.</p>
<p>While tools like AI-powered biometric verification help detect fraudulent video, they’re in a constant race with the same AI-technology that is used against them.</p>
<p>Safas goes beyond image analysis, and instead puts guard rails in place at the root of it all: By certifying that all video in a call originates from real camera hardware in real-time, blocking non-genuine feeds from virtual cameras, including any recordings and deepfakes, right at the source.ktop and web applications.</p>
<p>Safas is currently in closed beta.  <a href="/waitlist">Join our waitlist</a> for a trial.</p>

Top 5 Emerging Security Threats for Video Conferencing in 2025

<p>In recent years, AI-generated content has evolved to the point where deepfake technology has materialized from theory to a major threat in virtual meetings. Key advancements in AI-generated content have fueled this rise in impersonation attempts, particularly in the professional world, where sensitive communications often occur over desktop-based video calls. Unfortunately, these same advancements have enabled cybercriminals to bypass traditional security measures with alarming ease:
Whether it’s a fraudster impersonating an executive to defraud a subordinate for financial gain, or simulating an employee to access sensitive company information, the applications of this technology are are diverse. Financial fraud and data breaches are some of the more dangerous consequences, with bad actors able to trick their targets using lifelike simulations of real people.</p>
<h2>Identity Verification: A Partial Solution</h2>
<p>The B2B world’s most common proposal to tackle this growing problem is to embed more biometric identity verification (IDV). Technology such as Apple’s Face ID leverages machine learning and facial biometrics to verify the iPhone user’s facial biometrics before admitting access to the phone or processing a purchase. This approach relies on sophisticated algorithms that can match a person’s facial features to an image stored in a secure system.</p>
<p>And IDV technology is indeed highly effective for use cases that are deeply integrated with access to hardware itself, which is why most IDV will involve a mobile-based application such as when boarding your flight at the airport using your phone, or when using your mobile banking app. That is why fraudsters have shifted focus to the business world where the most vulnerable scenario is the most common: Holding virtual meetings via desktop- and browser-based video conferencing. In these scenarios, IDV cannot distinguish between a legitimate webcam feed and a virtual camera feed that is injecting pre-recorded video of the real person, undermining its effectiveness in these scenarios.</p>
<p>As most identity verification systems are optimized for mobile devices, platforms like Zoom or Google Meet—which are primarily used on desktops—remain vulnerable to these kinds of attacks. This has turned desktop-based virtual meetings into a primary target for attackers seeking to exploit the gap between mobile and desktop security measures.</p>
<h2>Unresolved Gaps: The Problem with Virtual Camera Injection</h2>
<p>By focusing on mobile-first solutions, companies are overlooking the inherent weaknesses of desktop environments. This oversight creates a dangerous blind spot that attackers can easily exploit: Desktop-based video conferencing platforms is video injection via a virtual camera. On commonly used B2B videoconferencing tools such as Zoom, Google Meet, and WebEx, free tools like OBS can be used to create a virtual camera that feeds any content in real time, pre-recorded or live-generated. The virtual camera is then chosen by the fraudster to replace their live webcam feed with the alternative content, which includes pre-recorded video or even real-time deepfakes, making it difficult for meeting participants to discern whether they are speaking to a real person or an AI-generated doppelgaenger.</p>
<p>These vulnerabilities highlight the need for stronger verification methods within browser-based and desktop videoconferencing environments. Unfortunately, neither IDV nor videoconferencing apps like Zoom or Microsoft Teams are equipped to detect the use of virtual cameras in their browser-based meetings, and there is no way to understand whether a participant is joining from a browser or from the web application. The fact that a fraudster would target and exploit this well-known vulnerability effectively compromises the integrity of all business meetings, as proven with documented attacks of <a href="https://www.cnn.com/2024/05/16/tech/arup-deepfake-scam-loss-hong-kong-intl-hnk/index.html" rel="noopener noreferrer" target="_blank">deepfake-impersonations in meetings that have caused millions of dollars in losses.</a></p>
<h2>The Future of Secure Videoconferencing: Certified Hardware Streams</h2>
<p>To address these vulnerabilities, the team at Safas came up with an unconventional solution: certified hardware streams. This technology guarantees the integrity of video feeds by ensuring that the content captured by the camera is exactly what is displayed during the video conference. By securing the video stream at the hardware level, it becomes impossible for attackers to swap out real-time footage with pre-recorded video or deepfakes through virtual camera injection.</p>
<p>This approach ensures that only authenticated, unmanipulated content is streamed during meetings, providing peace of mind to organizations concerned about the security of their virtual communications.</p>
<h2>Conclusion: Securing the Future of Videoconferencing</h2>
<p>As AI-generated content and deepfake technology continue to evolve, businesses must adapt to secure their everyday virtual communications at the root: By ensuring the integrity of hardware-streams, organizations can fortify the security of all their videoconferencing interactions. Certified authentic camera feeds are foundational for preventing real-time deepfakes and ensuring that what you see is what the camera captured, even on typically vulnerable desktop and web applications.
Safas is still in closed stealth beta stage. <a href="/waitlist">Join the waitlist</a> now to receive updates on when Safas becomes available, and be among the first to be contacted.</p>

Always know who you’re

really

talking to.

Trust in real-time video is eroding, accelerated by AI's impact and the people determined to abuse it for impersonation.

We counter this trend with technology that brings back a non-negotiable: The need to know that what you see on screen is real and uncompromised.

WHO WE ARE AND WHAT WE DO

A need to keep it real (again).

Latest Blog Posts

Deep-Live-Cam: High-quality deepfake avatars, free for anyone

Top 5 Emerging Security Threats for Video Conferencing in 2025

Real-Time AI-Generated Deepfakes are here: An Imminent Threat for Virtual Meetings

News

Meta’s new “Movie Gen” AI system can deepfake video from a single photo

GPT-4o: OpenAI’s shield against $40B deepfake threat to enterprises

Deepfake Ukrainian diplomat targeted US senator on Zoom call

Do you want to know more?