Google Launches Voice Spoofing Detection and New AI Tools for Creating Images and Videos

Google has released a new safety feature for Android phones that can detect when someone is using AI to fake a caller's voice, alongside updated tools for generating videos and images. The moves address both protecting users from AI-based threats and enabling creative uses of the same technology.

Android Gets Protection Against AI Voice Cloning

The new Android feature can spot calls where someone has used AI to copy another person's voice. According to Google's announcement, this is one of the first tools of its kind designed to catch voice deepfakes — AI-generated audio that mimics someone's speech patterns so convincingly it tricks you into thinking they're the person calling.

The feature works when both the person calling and the person receiving the call use Phone by Google, Google's calling app. When a suspicious call comes in, the system analyzes the voice and compares it to what it knows about the person who claims to be calling. If something doesn't match up, it flags the call as potentially fake.

The timing matters. Voice cloning AI has gotten much better and easier to use. A few years ago, this technology needed hours of recorded speech to work well. Now it can create a convincing voice fake from just minutes of audio. That means scammers and criminals can impersonate trusted contacts — your boss, a family member, a bank representative — more easily than ever, which is why security experts have been pushing tech companies to develop defenses.

Veo 2 and Updated Imagen 3 Join the Creative AI Lineup

Google is releasing Veo 2, a newer version of its video-generation AI, along with an updated version of Imagen 3, its tool for turning written descriptions into images. Both are available through Google Labs, which is Google's testing ground for experimental features. You can try them through tools called VideoFX and ImageFX, and Google has also introduced a new experiment called Whisk.

Veo 2 is Google's answer to OpenAI's Sora and other AI video generators. The key differences between these tools usually come down to how fast they work and how good the videos actually look. Imagen 3 competes with other image generators like Midjourney and DALL-E. With these kinds of tools, what keeps people using one over another is usually how good the final images are and how well the AI understands what you're asking it to create.

What Is Whisk?

Google Labs has introduced Whisk, described as a new experiment in image creation. The details are still sketchy, but the company's framing it as an "experiment" suggests they're testing a different way of creating images than the standard approach of just typing a description and getting a result back.

All three of these tools — Veo 2, Imagen 3, and Whisk — now live in one place within Google Labs. This means you don't have to jump between different apps or websites to use them. It's the kind of consolidation that makes a product easier to use.

Google is pulling off something that sounds contradictory: it's building defenses against AI-powered threats while simultaneously creating the very generative AI tools that could be misused. The fake call detection is defensive; Veo 2 and Imagen 3 are generative. Both are necessary. You can't ignore either one.

The speed of this technology shift has been striking. I covered the early days of computer-generated speech in the early 2010s, when making a voice sound human took serious processing power. Back then, a natural-sounding synthetic voice was impressive but clearly artificial. Over the last three years, the jump from basic AI voices to convincing fakes has happened much faster than that earlier evolution did. The acceleration is real.

Looking at how this works under the hood, the fake call detection probably listens to incoming audio and compares it against a stored profile of what the real caller's voice sounds like. The system may use encryption or shared verification between the caller and receiver rather than relying only on audio analysis. That approach would catch more cases and produce fewer false alarms — important when a genuinely wrong caller ID could cause real harm.

For the image and video generators, keeping them inside Google Labs for now lets Google watch how people use them, improve them based on what it learns, and avoid some of the misuse problems that can emerge when a powerful tool reaches everyone at once. This is a more thoughtful approach than the anything-goes release strategies that marked earlier phases of the current AI boom.

This announcement also signals that Google's security team and AI research team are working together — at least in this case — rather than ignoring each other. That's a sign of maturity in how the company approaches AI deployment.

As voice synthesis tools become easier to access and cheaper to run, other companies will face the same pressure. Apple, Microsoft, and telecom carriers all need to solve this problem too. The fake call detection may seem like a narrow feature, but it's likely to become table stakes.

Similarly, improvements to Veo and Imagen will shape which creative AI platform people choose. Google's approach of bundling these tools inside Labs gives it a way to test features before rolling them out more broadly to products like Search, Gmail, and YouTube.

The fact that Google is releasing both defensive and generative AI tools at once reflects a hard-won lesson: these technologies are linked. You can't build one responsibly without building the other. As AI capabilities advance, the security infrastructure protecting against misuse has to advance at the same pace, or users lose confidence in the platform.