Featured inBloombergThe Wall Street JournalThe Telegraph
Chad, Microsoft's European CTO for AI and Copilot, invited Matthew to give the opening keynote at a closed-door event in Geneva, where a small, select group of Microsoft clients met to explore the future of AI, and in particular deepfakes, disinformation, misinformation and synthetic content.
It was familiar ground. Matthew has tracked the space for over a decade, was one of the first to identify the generative AI and deepfake trend in 2016, and wrote a book on it in 2019, three years before ChatGPT made the topic top of mind for everyone.
"I don't believe we will ever be able to catch all of the AI generated or manipulated content, even with the best tools. From here on in we should not believe anything behind a glass screen."Matthew Griffin, Geneva, March 2025
In a twenty-minute talk, followed by an hours-long fireside chat, he set out the fundamentals: the cost, resources and skill needed to create authentic-looking, high-quality fake content are all dropping to zero. In the coming years, the only thing holding back a flood of deepfakes and malicious synthetic content will be detection tools like those Microsoft is developing. He also drew a clear line between deepfakes, which manipulate original content with AI, and synthetic content, which AI generates from scratch, whether audio, images, VFX, video, text or virtual reality assets. The two have distinct upsides, downsides, characteristics and ecosystems.
Matthew closed by arguing that, from a disinformation perspective, we will never catch all AI-generated or manipulated content, even with the best tools. He demonstrated the point live, using simple techniques to evade Microsoft's own AI deepfake detection tools. His conclusion for society: from here on, we should not believe anything behind a glass screen.
311 Intelligence
What this means for your leadership team
Assume detection will miss some fakes.
Matthew evaded detection tools live using simple techniques. Processes must hold up even when a fake gets through.
Verify through a second channel.
If nothing behind a glass screen can be trusted, important requests need confirming by another route.
Treat deepfakes and synthetic content separately.
They have different upsides, downsides and ecosystems, so policies for one will not cover the other.
Three questions for your next leadership meeting
Which of our decisions could a convincing fake voice or video trigger?
How do we confirm high value requests made by phone or video?
Where are we already using synthetic content, and how do we label it?
The trends and technologies behind this session
From the 311 Starburst, the 311 Strategic Intelligence Unit’s map of the technologies and trends shaping the next 50 years.
Matthew showed that simple techniques can evade even advanced detection tools, so some fake content will always get through.
How widespread is synthetic content?
The 311 Starburst tracks AI made text, images, music and video fast becoming a large share of everything published online, and rates it Act now.
What is synthetic voice detection?
Software that listens to a phone or video call and warns if the voice is fake or cloned by AI. The 311 Starburst expects wide use around 2028.
How can 311 brief our leadership team on this?
Through a private executive briefing or leadership workshop, tailored to your sector, in person first.
Bring this to your leadership teamA private executive briefing or leadership workshop, tailored to your sector. In person first, anywhere in the world.
Analysis by the 311 Strategic Intelligence Unit, from Matthew Griffin’s session and 311 data.
If Matthew gave this session today
Bots now make 53% of web traffic.
The Dead Internet Theory, Priced finds bots make 53% of web traffic, yet people still do the buying. When fakes cannot all be caught, knowing what is real and reaching real customers matters more than ever.
