Blog Details

An audio deepfake impersonating Secretary of Recount Marco Rubio contacted international ministers, a U.S. governor, and a member of Congress with AI-generated voicemails mimicking his allege, in accordance to a senior U.S. legitimate and a Recount Division cable dated July 3. 

There’s no public evidence that any of the recipients of the messages, reportedly designed to extract sensitive info or form memoir entry, were fooled by the scam. However the incident is the most up-to-date high-profile example of how easy—and alarmingly convincing—AI allege scams become.

With appropriate 15 to 30 seconds of any individual’s speech uploaded to services admire Eleven Labs, Speechify and Respeecher, it’s now that it is doubtless you’ll imagine to form out any message and bear it read aloud of their allege. Withhold in solutions, these are instruments oldschool completely legitimately for a host of things from accessibility to direct advent – nevertheless admire many AI technologies, would perhaps perchance presumably be misused by depraved actors.

The specter of deepfakes has escalated

AI-generated deepfakes aren’t mute, particularly of C-suite leaders and public officials, nevertheless they’re turning valid into a better area. Eight months within the past, I reported that extra than half of chief info security officers (CISOs) surveyed ranked video and audio deepfakes as a growing trouble. That threat has handiest escalated. A brand mute survey by Surfshark found that within the indispensable half of 2025 on my own, deepfake-related incidents surged to 580—nearly four conditions as many as in all of 2024 (150 incidents), and dramatically greater than the 64 incidents reported between 2017 and 2023. Losses from deepfake fraud bear additionally skyrocketed, reaching $897 million cumulatively, with $410 million of that in precisely the indispensable half of 2025. The commonest blueprint: impersonating public figures to advertise false investments, which has already resulted in $401 million in losses.

“Deepfakes bear developed into valid, energetic cybersecurity threats,” Aviad Mizrachi, CTO and cofounder of map security firm Frontegg, told me by email. “We’re already seeing AI-generated video calls successfully trick workers into authorizing multimillion-dollar funds. These assaults are occurring now, and it’s a scam that’s turning into alarmingly easy for a hacker to deploy.”

Segment of the world, Mizrachi added, is that oldschool authentication programs—usernames, passwords, one-time codes, and authenticator apps—weren’t designed for a world where a scammer can clone your allege or face in seconds. That’s attributable to these scams don’t necessarily hold breaking into an memoir—they rely on tricking a valid particular person into handing over credentials or authorizing actions themselves.

“These oldschool safety features to take a look at the identity of an particular person obviously don’t work anymore,” he acknowledged, including that the majority cybersecurity groups quiet miss out on deepfakes—and that’s the vulnerability attackers exploit. A convincing fake allege on a allege mail message or video name can persuade any individual to circumvent traditional procedures or approve a wire switch, despite the truth that the total authentication instruments are technically in space.

To guard against that extra or much less deception, Mizrachi acknowledged, organizations need to deploy stronger security instruments that rely on physical devices—admire a smartphone or hardware security key—to show any individual’s identity. These instruments, is known as FIDO2 or WebAuthn passkeys, are a ways more challenging for hackers to fake or phish. And former instrument checks, trim verification techniques can additionally music behavioral alerts—admire typing tempo, residing, or login habits—to residing anomalies that a cloned allege can’t imitate. These extra layers execute it much more challenging for a deepfake attack to be triumphant.

Margaret Cunningham, director of security and AI strategy at security firm Darktrace, acknowledged that the impersonation strive of Rubio demonstrates appropriate how with out drawback generative AI would perhaps perchance presumably be oldschool to launch convincing, centered social engineering assaults.

“This threat didn’t fail attributable to it become poorly crafted—it failed attributable to it left out the ethical second of human vulnerability,” she acknowledged. “Of us most steadily don’t execute choices in aloof, centered conditions. They retort whereas multitasking, below stress, and guided by what feels familiar. In these moments, a depended on allege or legitimate-searching message can with out drawback bypass warning.”

Deepfakes bear impacted democracies spherical the field

Generative AI has additionally dramatically reduced the barrier to entry just about media manipulation—making it sooner, more inexpensive, and extra scalable than ever before. And it is impacting democracies spherical the field: A mute Contemporary York Times file found that AI-powered deepfakes has reworked elections in as a minimum 50 countries spherical the field, by the usage of them to demean or defame opponents. 

“Here is the mute frontier for influence operations,” Leah Siskind, AI analysis fellow at the Foundation for Protection of Democracies, told me. “We’ve viewed completely different cases of deepfakes of senior authorities officials oldschool to form entry to non-public accounts, nevertheless leveraging AI to influence diplomatic relationships and possibility-making is a foul escalation. Here is an urgent nationwide security area with serious diplomatic ramifications.”

For now, Siskind recommends that authorities officials steer definite of accepted encrypted platforms admire Signal, which whereas rep in terms of direct, lack mechanisms for identity verification. “Given the ease of constructing deepfake audio and building out real looking-searching accounts on any particular person-grade messaging app, senior authorities officials need to quiet follow rep dialog channels,” she acknowledged.

Show: Inspect this mute Fortune video about my tour of IBM’s quantum computing take a look at lab. I had a amazing time inserting out at IBM’s Yorktown Heights campus (a midcentury stylish wonder designed by Eero Saarinen, the identical guy because the St. Louis Arch and the classic TWA Flight Heart at JFK Airport) in Contemporary York. The video become fragment of my coverage for this year’s Fortune 500 area that incorporated an editorial that dug deep into IBM’s mute rebound.

As I acknowledged in my fragment, “walking thru the IBM analysis heart is admire stepping into two worlds immediately. There are the steel and glass curves of Saarinen’s invent, punctuated by wide partitions fabricated from stones aloof from the encompassing fields, with long-established Eames chairs dotting discussion nooks. However this twentieth-century modernism contrasts starkly with the graceful, wide, fridge-admire quantum laptop—among the many most progressed on this planet—that anchors the collaboration house and dealing lab, where it whooshes with the regular hum of its cooling machine.”

With that, here’s the remainder of the AI info.

Sharon Goldman
sharon.goldman@fortune.com
@sharongoldman

AI IN THE NEWS

xAI releases Grok 4 amid backlash over antisemitic posts. Elon Musk’s xAI has launched Grok 4, appropriate months after its old launch, highlighting the quick tempo of AI vogue. Unveiled at some stage in a gradual-night livestream, Musk claimed the mute chatbot outperforms most graduate college students across disciplines and now helps improved allege interactions. xAI additionally touted benchmark results showing Grok 4 beating opponents admire OpenAI. Whereas Musk admitted the bot most steadily lacks classic sense and hasn’t but made scientific breakthroughs, he added, “that’s appropriate a topic of time.” Grok 4’s launch comes appropriate a day after xAI faced backlash for antisemitic direct generated by the chatbot on X. The firm eliminated the offensive posts and acknowledged it has since taken steps to block detest speech before Grok’s responses are printed on the platform.

Perplexity launches AI browser to clutch on Google—starting with its vitality customers. Perplexity is stepping up its challenge to Google with the launch of Comet, its first AI-powered internet browser. Accessible in the starting up to $200/month Max subscribers and a restricted waitlist, Comet integrates Perplexity’s signature AI search engine front and heart—providing summarized solutions barely than oldschool search results. The browser additionally debuts Comet Assistant, an in-browser AI agent that can summarize emails and calendar occasions, residing up tabs, and interact with webpages in valid time. CEO Aravind Srinivas has framed Comet as extra than appropriate a browser—it’s a stepping stone to what he calls an “AI working machine” designed to deeply embed Perplexity into customers’ day after day workflows. The switch places Perplexity in extra convey competition with Chrome and Google’s enjoy AI search experiments, because the startup bets on its browser turning into the mute gateway for how of us earn and interact with info online.

OpenAI is reportedly gearing up to launch its enjoy internet browser. In accordance with Reuters, OpenAI is additionally becoming a member of the browser wars because it moves to compete extra without prolong with Google, and now Perplexity. Citing three sources conversant within the plans, the file says the browser would perhaps perchance debut within the upcoming weeks and is designed to retain some particular person interactions internal a ChatGPT-model interface—barely than directing customers to external internet sites. The switch would additionally give OpenAI entry to the extra or much less particular person info that has prolonged powered Google’s dominance in search.

FORTUNE ON AI

Apple’s AI efforts ‘bear struck hour of darkness’ and the handiest capacity it’ll end getting extra at the abet of is shopping Perplexity, analyst Dan Ives says —by Marco Quiroz-Guitierrez

Would you exchange your CEO with an AI avatar? —by Alexandra Sternlicht

Why Coca-Cola’s CIO prioritizes tall-influence AI pilot initiatives —by John Kell

Amazon’s tariff-clouded, vendor-careworn, AI-researched, most unearthly Top Day ever —by Jason Del Rey

AI CALENDAR

July 13-19: Global Convention on Machine Finding out (ICML), Vancouver

July 22-23: Fortune Brainstorm AI Singapore. Prepare to attend here.

July 26-28: World Man made Intelligence Convention (WAIC), Shanghai. 

Sept. 8-10: Fortune Brainstorm Tech, Park City, Utah. Prepare to attend here.

Oct. 6-10: World AI Week, Amsterdam

Dec. 2-7: NeurIPS, San Diego

Dec. 8-9: Fortune Brainstorm AI San Francisco. Prepare to attend here.

EYE ON AI NUMBERS

Data breach 80%

That’s how much industry map will be powered by AI that understands extra than appropriate text—including photos, video, audio, and completely different info—by 2030, in accordance to a mute file from Gartner. That’s a immense jump from lower than 10% this day.

This shift, is known as “multimodal AI,” would perhaps perchance exchange how companies feature across industries admire healthcare, finance, and manufacturing. As an example, it’ll also abet map execute smarter choices by analyzing a combination of information (admire medical photos and affected person notes), or even clutch proactive steps—admire flagging fraud or optimizing provide chains—with out human input.

Gartner analysts scream companies building map would possibly want to launch investing now in these mute AI technologies to live competitive and ship valid cost to their possibilities.