As democratic elections in more than forty countries approach, twenty of the world's most powerful technology companies have gathered under a voluntary accord, pledging to detect and diminish the influence of AI-generated deepfakes on political life. The agreement arrives not as a precaution but as a response — synthetic voices have already reached tens of thousands of voters, and fabricated images have already circulated in campaign materials. What remains unresolved is the oldest tension in self-governance: whether those who hold power will constrain themselves without being compelled to do
20 Tech Giants Pledge to Combat Election Deepfakes, But Enforcement Remains Unclear
AI-generated deepfakes of political candidates and robocalls impersonating President Biden reached thousands of voters, potentially influencing electoral participation.
Whether voluntary commitments will prove sufficient against coordinated disinformation campaigns remains an open question.
Mark
So twenty companies signed this thing on Friday. What exactly did they promise to do?
Mimi
They committed to eight things, basically: develop tools to detect and remove deceptive AI content from their platforms, assess their own models for election risks, share detection technology across the industry, be transparent about how they handle it, and run educational campaigns about media literacy.
Luke
But here's the thing—every single commitment uses soft language. "Seek to detect." "Work together." "Provide transparency." There's no enforcement mechanism. If a company doesn't follow through, what happens?
Mimi
That's the gap everyone's pointing out. Rachel Orey from the Bipartisan Policy Center said the language isn't as strong as you'd hope, and it's all voluntary.
Mark
Has this actually been a problem yet? Are deepfakes already being used in elections?
Mimi
Yes. In January, someone used ElevenLabs' voice-cloning tool to create a fake Biden robocall that reached 25,000 voters in New Hampshire, telling them not to vote. The RNC and DeSantis campaign both used AI-generated images in their ads last year.
Luke
Wait—ElevenLabs is one of the twenty companies that signed this accord. So they're committing to stop the very thing their own tool was used for.
Mimi
Right. Though to be fair, they say they're implementing safeguards now. OpenAI is watermarking images, preventing chatbots from impersonating candidates.
Mark
Who didn't sign?
Mimi
Midjourney, which makes some of the most convincing fake images out there. And Apple, though that's probably because they haven't launched generative AI products yet.
Luke
So the company with arguably the most powerful image generator isn't bound by any of this. And the accord has no way to enforce compliance anyway.
Mark
What's the actual risk here if this doesn't work?
Mimi
Coordinated disinformation campaigns using AI to sway voters in elections across forty-plus countries. Fake videos of candidates, fake audio, false voting information—all at scale and speed that's hard to counter.
Luke
The FCC did ban AI robocalls this month, so there's at least some regulatory action happening. But Congress hasn't passed any AI legislation, and the EU's AI Act is still being implemented.
Mark
So we're relying on companies to police themselves while the rules are still being written.
Mimi
For now, yes. And we'll find out over the next several months whether these commitments actually mean anything.
El Pulso
- AI-generated deepfakes are no longer a future threat — a cloned Biden voice reached 25,000 New Hampshire voters urging them to stay home, and synthetic campaign imagery has already aired from major political operations.
- Twenty tech giants, including OpenAI, Google, Meta, and ElevenLabs — the very company whose voice-cloning tool enabled the Biden robocall — signed a collective accord pledging to detect, remove, and be transparent about deceptive AI election content.
- The accord's language is entirely voluntary, with no penalties for non-compliance, leaving critics and election integrity experts skeptical that pledges alone can outpace coordinated disinformation at scale.
- Notable absences sharpen the concern: Midjourney, producer of some of the most photorealistic synthetic imagery available, declined to sign, while Apple and unaffiliated open-source tools remain outside the coalition's reach.
- The EU has already enacted binding AI legislation, the FCC has banned AI robocalls in campaign contexts, but the US Congress has passed nothing — leaving this industry pact as the primary American bulwark heading into a consequential election year.
As democratic elections in more than forty countries approach, twenty of the world's most powerful technology companies have gathered under a voluntary accord, pledging to detect and diminish the influence of AI-generated deepfakes on political life. The agreement arrives not as a precaution but as a response — synthetic voices have already reached tens of thousands of voters, and fabricated images have already circulated in campaign materials. What remains unresolved is the oldest tension in self-governance: whether those who hold power will constrain themselves without being compelled to do so.
On a Friday in February, twenty technology companies — among them OpenAI, Google, Meta, Amazon, Adobe, and X — signed a voluntary accord committing to fight AI-generated deepfakes ahead of elections in more than forty countries. The signatories pledged to build detection tools, audit their own models for election-related risks, remove deceptive content, and support public media literacy. The agreement covers synthetic audio, video, and images designed to impersonate candidates, election officials, or misrepresent voting procedures.
The accord's eight commitments, however, carry no enforcement mechanism. Companies will "seek" to detect content and "work together" on tools — language that is aspirational rather than binding. Rachel Orey of the Bipartisan Policy Center acknowledged the effort while noting its central weakness: follow-through is entirely voluntary, and consequences for non-compliance do not exist.
The urgency is not hypothetical. The 2024 cycle has already seen AI-generated images in RNC and DeSantis campaign ads, and in January, two Texas firms used ElevenLabs' voice-cloning technology to impersonate President Biden in robocalls sent to as many as 25,000 New Hampshire voters, urging them to skip the primary. The FCC responded with a unanimous ban on AI-generated robocalls in campaign contexts. ElevenLabs, notably, is among the accord's signatories.
OpenAI announced concrete countermeasures alongside the accord: DALL-E 3 images will carry digital watermarks, and its chatbots will be prevented from impersonating candidates. Microsoft's Brad Smith framed the industry's responsibility plainly — AI did not invent election deception, but it must not be allowed to amplify it.
Midjourney, whose generator produces some of the most photorealistic synthetic imagery available, declined to join. Stability AI signed on; Apple, which has released no generative AI products, did not participate. The European Union has already passed a comprehensive AI Act with binding safety requirements, while the US Congress has enacted nothing. The accord stands, for now, as the industry's attempt to self-regulate before governments intervene — a wager whose odds will be tested across a year of consequential elections.
Twenty technology companies put their names to a voluntary accord on Friday aimed at stopping AI-generated deepfakes from poisoning elections in more than forty countries during 2024. The signatories—OpenAI, Google, Meta, Amazon, Adobe, X, and fifteen others—committed to developing detection tools, assessing their own models for election-related risks, and working to remove deceptive content from their platforms. They also pledged to be transparent with the public about how they handle such material and to support media literacy efforts. The agreement covers AI-generated audio, video, and images designed to fake the appearance or voice of political candidates, election officials, or other key figures, as well as false information about voting procedures.
Yet the accord's eight commitments, while broadly stated, contain no binding enforcement mechanism. The language is voluntary throughout—companies will "seek" to detect content, will "work together" to create tools, will "provide transparency." Rachel Orey, senior associate director of the Elections Project at the Bipartisan Policy Center, acknowledged the effort but flagged the weakness: the agreement is voluntary, and whether companies actually follow through remains to be seen. The question hanging over the pact is whether a loose coalition of pledges, without teeth or consequences for non-compliance, can meaningfully slow the weaponization of generative AI in democratic contests.
Deepfakes have already entered the 2024 cycle. In April 2023, the Republican National Committee released campaign advertisements featuring AI-generated images of President Joe Biden and Vice President Kamala Harris. That June, the Ron DeSantis campaign followed with synthetic images of Donald Trump, his rival for the GOP nomination. Both included disclaimers about the AI origin, though they were easy to overlook. More troubling was the robocall campaign in January: two Texas-based companies used ElevenLabs' voice-cloning technology to generate an impersonation of Biden's voice, then deployed it via robocall to as many as 25,000 voters in New Hampshire, urging them not to vote in the state's primary on January 23. The Federal Communications Commission responded swiftly, voting unanimously earlier this month to ban AI-generated robocalls in campaign contexts.
ElevenLabs, the company whose technology enabled the New Hampshire robocalls, is among the twenty signatories to the accord. So is OpenAI, which has already announced its own countermeasures: images generated with DALL-E 3 will carry a digital watermark classifier marking them as AI-created, and the company plans to work with journalists, researchers, and platforms to refine its detection systems. OpenAI also said it would prevent its chatbots from impersonating candidates. Microsoft's Brad Smith, in the accord's press release, framed the commitment as a responsibility: AI did not invent election deception, but the industry must ensure it does not amplify it.
Notably absent from the list is Midjourney, whose image generator produces some of the most photorealistic synthetic pictures currently available—the company famously created a viral fake image of Pope Benedict XVI in a puffy white jacket last year. Midjourney said earlier this month it would consider banning political image generation during election season, but it did not join the accord. Stability AI, Midjourney's main competitor and the maker of the open-source Stable Diffusion model, did sign on. Apple, the only member of Silicon Valley's Big Five not to participate, has not yet released generative AI products and does not operate a social media platform where deepfakes would circulate, which may explain its absence.
The European Union has moved faster on regulation: in December, it agreed on a comprehensive AI Act with safety requirements that could influence other nations' approaches. The US Congress, by contrast, has not passed any AI legislation. The accord represents an attempt by the industry to self-regulate before governments impose rules, but whether voluntary commitments will prove sufficient against coordinated disinformation campaigns remains an open question as the election year accelerates.
Citas Notables
The language isn't quite as strong as one might have expected. It is voluntary, and we'll be keeping an eye on whether they follow through.— Rachel Orey, senior associate director of the Elections Project at the Bipartisan Policy Center
AI didn't create election deception, but we must ensure it doesn't help deception flourish.— Brad Smith, Microsoft Vice Chair and President
Quieres la nota completa?
Lee el original en Engadget ↗
Technology