AI Cybersecurity Risks Heading Toward the 2028 Elections

Federal warnings, researchers and AI executives point to the same problem: deception got faster and cheaper, while voter and campaign defenses did not keep pace.

By PetaTech24 Technology Desk, October 25, 2026

The AI cybersecurity risks facing American elections are no longer a distant worry for 2028. They are already visible in the 2026 midterms, in federal warnings issued in October 2026, and in public arguments between AI companies and lawmakers about safety.

One transparency note matters first. PetaTech24 could not open the original Axios page, so this article does not quote that piece directly. It builds on corroborating coverage: a Reuters report on a joint FBI and CISA warning from October 16, 2026, Brennan Center testing, and secondary reporting on Axios coverage.

That distinction matters because the 2028 debate mixes two arguments. One is about cybersecurity in the narrow sense: phishing, hacked accounts and disrupted systems. The other is about trust: fake videos and forged documents that make real results harder to believe.

The Short Version

• The FBI and CISA warned on October 16, 2026 that election officials, candidates and political groups are attractive targets for AI-enhanced phishing by text and email.
• Brennan Center researchers found that popular AI tools could be pushed to generate convincing election misinformation at scale, including fake images and forged-looking documents.
• Secondary reporting on Axios coverage says AI executives have rehearsed responses to a major AI-linked incident, while lawmakers in both parties press companies to focus on product safety.
• The AI cybersecurity risks for 2028 center less on changing vote totals, which researchers describe as unlikely to go undetected, and more on eroding confidence before and after voting day.
• Voters can cut their own exposure now: verify unexpected messages through a second channel, distrust viral election claims without an official source, and check local election office sites directly.

The Bigger Picture

AI policy is now election politics, not just technology policy. Axios reported on October 14, 2026 that AI safety groups and donors spent $48 million this election cycle on Democratic and Republican races, according to an analysis summarized by OpenTools, which cautions that the figure is Axios’s own tally across different entities, not a single committee total.

According to Aju Press’s account of Axios reporting from October 24, White House officials and federal lawmakers, including some who support AI development, urged AI company leaders to stop talking publicly about AI’s downsides and focus on product safety.

The safety debate inside the companies runs in parallel with the AI cybersecurity risks debate in Washington. Summaries of Axios reporting by Forklog and OpenX say executives at Anthropic, OpenAI and other AI firms privately rehearsed responses to a major AI-linked incident, with a large cyberattack on banking, internet access, power or water named as a feared scenario. Insiders cited by Axios expect a major incident within six to 12 months, while OpenAI said its exercises cover scenarios not treated as inevitable.

AI Cybersecurity Risks in the 2026 Midterms

The most concrete warning so far came from the government, not the industry. Reuters reported on October 16, 2026 that the FBI and CISA urged election officials and the public to guard against AI-powered tricks ahead of the November midterms, saying hackers may use bogus messages to steal passwords and that AI makes those messages more convincing. This is one of the clearest examples of the AI cybersecurity risks voters face.

That warning is narrower than it first sounds. It does not claim voting systems were compromised. It says people around elections are targets: officials, candidates and their political organizations.

Brennan Center researchers tested what AI tools will generate when pushed. In work published in September 2026 and updated in October, they tested ChatGPT, Gemini, Grok, Meta AI, Runway and Flux.2. They report that all four chatbots questioned helped frame election misinformation themes, only Grok produced their full target of 100 image-generation prompts, and all six tools then generated images from those prompts that could spread election misinformation.

Foreign influence operations add a second layer. The Brennan Center reports that Russia, China and Iran are deploying AI tools in more sophisticated influence operations aimed at the United States, and cites reporting that Chinese actors stood up at least 5,000 inauthentic X accounts controlled by an AI language model system.

Separate reporting on autonomous AI agents points to the next step in scale. Particle News summarized warnings in October 2026 that agents can run disinformation campaigns with far fewer people than past troll operations. Experts assess direct manipulation of vote totals as unlikely, but warn that agents could erode confidence.

What the Warnings Actually Cover

Sorting the warnings into three buckets helps. Phishing is a hygiene problem: strong authentication and verified contact channels. Synthetic media is a verification problem: provenance labels and checking the original source. A large infrastructure attack is a resilience problem: backup systems and rules decided before a crisis. Each bucket is part of the wider AI cybersecurity risks picture.

On synthetic media, the policy response is already partly law. The Brennan Center notes that the EU AI Act Article 50 and the California AI Transparency Act require marking of AI-generated content, with provenance data required as of August 2, 2026. California is due to require social media labels from 2027, and device makers to offer digital signing of authentic captures from 2028.

AI Cybersecurity Risks to Watch Before 2028

Three developments deserve attention between now and 2028. First, whether provenance and labeling rules actually change what voters see, or whether unlabeled copies circulate faster than labeled originals. California’s 2027 labeling step and 2028 device signing step will be the first large United States test of that idea.

Second, whether federal election security capacity is rebuilt or keeps shrinking. The Brennan Center argues that rebuilding federal capabilities to detect foreign misinformation campaigns will likely have to wait for a new president and Congress.

Third, whether AI companies treat election safeguards as a shared standard or a competitive choice. The Brennan Center tests show uneven behavior: some tools refused certain prompt-writing requests while others complied, yet all generated the resulting images. Consistency across models, plus independent testing, is the test to watch, and one of the clearest AI cybersecurity risks on the road to 2028.

For ordinary voters, the advice stays unglamorous and effective. Use hardware keys or app-based authentication where offered, and never reset a password from a link in an unexpected message. Check voting dates and results only on your state or local election office website, not from a screenshot.

Quick Answers

Will AI change vote totals in 2028? Researchers cited here assess direct manipulation of vote totals as unlikely to go undetected, partly because paper records and local checks still exist. The larger AI cybersecurity risks concern false evidence and lost confidence.

Can you spot every deepfake? No, and that is the point of the Brennan Center findings on the AI cybersecurity risks from synthetic media. Realistic fakes can be generated on the first try, and watermarks can be removed. Provenance data and original-source checks help, but no single signal is reliable alone.

What should you do with a suspicious election claim? Pause before sharing, search for it on your local election office site or a reputable news outlet, and check whether the original video or document can be found from its claimed source.

The Takeaway

The AI cybersecurity risks heading toward 2028 are not one threat but three: phishing that targets people, synthetic media that targets belief, and the political fallout if a major cyber incident lands during an election cycle. The first two are documented in 2026. The third remains a rehearsed scenario, not a certainty.

What voters, campaigns and platforms do in the next two years will decide how much of this risk turns into real damage. Strong authentication, consistent safeguards across AI models, provenance labels that travel with content, and local election offices that publish clear, early information are all available now. Together they are the best available answer to the AI cybersecurity risks described here.

Sources: Reuters, October 16, 2026; Brennan Center, The Second AI Election, September 23, 2026, updated October 20, 2026; Aju Press, October 25, 2026; Forklog and OpenX summaries of Axios reporting; OpenTools, October 14, 2026; Particle News, October 2026. External references: Brennan Center election safeguards analysis and Aju Press report on Axios AI safety coverage.