News

'If you don't have visibility, you can't understand what to protect'

When it comes to securing enterprise supply chains, now heavily infused with AI applications and agents, a software bill of materials (SBOM) no longer provides a complete inventory of all the components in the environment. Enter AI-BOMs.…

Your voter data could be used against you. A foreign intelligence service that wished to identify the family members of deployed military personnel could do so by cross-referencing public voter record data and social media posts. An employer who only wanted to hire employees with a specific political affiliation could do so by analyzing the primary ballot history of job applicants. An identity fraud ring seeking to open credit accounts in the names of other people could identify voters whose mail has been returned (via voter file suspense indicators) to take over those addresses using bogus change-of-address requests. These scenarios are possible thanks to the ability to link publicly available voter data to other data sets, according to Noah M. Kenney, founder of consultancy Digital 520. "I picked two different counties that kind of represented opposite ends of the spectrum," Kenney told The Register in a phone interview.  "In Texas, they hide a lot of information and then North Carolina makes a lot of it public in terms of the specific records. And what I was looking at specifically is if you go and merge this data set or link this data set with other data sets, how likely are you to be able to re-identify a person?" More than 25 years ago, research by Latanya Sweeney, currently a professor at Harvard, demonstrated that most of the US population (87 percent) could be identified with just three anonymous data points – a five-digit ZIP code, gender, and date of birth. Those results can be improved when combined with other data sets. And recent research has shown that the process of identifying people from seemingly anonymous data points becomes even easier with AI tools. In a research paper titled "Public Voting Records: A Record, or an Attack Surface?", Kenney describes how he analyzed public records from Travis County, Texas, and Robeson County, North Carolina to show that the adversarial scenarios cited above are practical with public data. The Texas file provides fewer data points than the North Carolina file, but the research suggests redaction doesn't make much of a difference in the re-identification scenarios evaluated. Table 1 — Disclosure regime comparison With the less detailed Texas info, Kenney was able to use a Python script to link the voter records to other public records like the Federal Election Commission's individual-contribution data. "We pulled 500 contribution records for ZIP 78704 (an Austin-core ZIP including South Congress and Travis Heights neighborhoods) from the 2024 cycle via the FEC OpenAPI on May 1, 2026," he explains in his paper.  "We de-duplicated to 181 unique contributors by exact match on (last name, first name, ZIP), and inner-joined to the voter file on the same key, no fuzzy matching, no nickname normalization, no suffix handling. Of the 181 contributors, 105 (58.01 percent) matched any voter record and 95 (52.49 percent) matched a uniquely-identifiable voter. Of the 105 matches, 74.3 percent had a non-trivial employer field in FEC." That 52 percent individual match rate for identifying individuals from voter rolls and FEC data, Kenney said, would be more like 90–95 percent using the kinds of tools commercial data brokers employ. The North Carolina voter dataset includes a phone number for the majority of voters. According to the paper, 88.53 percent of voters who have a phone number listed have a number that is unique within the county. As a result, external datasets containing phone numbers can be joined at a similar rate using this field as a key to narrow down and identify likely individuals. Among the report's other findings:  Name and ZIP code uniquely identify 95.81 percent of Texas voters and 87.79 percent of North Carolina voters. Among Travis County voters who have voted in 20 or more elections, 98.4 percent have a turnout pattern that is unique to them, making that data point a fingerprint. Texas' redaction of date of birth as a privacy measure is undermined by the publication of the voter registration data, which allows 28 percent of voters to be uniquely identified when combined with ZIP and gender. The Travis County voter file currently exposes 320 deployed military families through the publication of APO/FPO codes for military mailings. There's currently no comprehensive federal privacy law. While many states have privacy rules, there's a lot of variation. "Even within a specific state, most of the counties are individually handling these public records requests, so they all handle them differently across the country," said Kenney.  "Some of them, you can't get them. Some of them, you need an ID to get them. Some of them you have to go through a request process for public records or you have to pay for them. The two counties I used are both freely available. You can go and download zip files of them without even putting in an email address or your name from anywhere in the world." Kenney said that he believes that access controls represent a better answer than redacting certain data fields, pointing to his findings that show redaction doesn't necessarily protect against privacy harms. He recommends measures like rate limits on bulk file requests, identity verification, requiring state ID, maintaining audit logs of requests, and prohibiting commercial resale of these records – because they're often used by data brokers. Beyond specific fixes based on his findings – Texas should generalize voter registration dates to a year rather than a day and armed forces mailing codes should be excluded from voter rolls – Kenney argues that people should be allowed to opt out of inclusion in public data sets and that general data privacy protections would be helpful. Last week, House Republicans introduced the Secure Data Act in an effort to create federal privacy rules. But Kenney says that it's significantly weaker than a lot of state regulations and he doesn't expect it will pass. "The industry consensus is that the likelihood of it passing is extremely low, at least in its current form," he said. "This represents the third attempt to pass comprehensive data privacy in recent years, most recent being the American Data Privacy and Protection Act, which failed to pass." ®

Even limited voter rolls can be linked to identify people, research shows

Your voter data could be used against you. A foreign intelligence service that wished to identify the family members of deployed military personnel could do so by cross-referencing public voter record data and social media posts.…

Information security agencies from the nations of the Five Eyes security alliance have co-authored guidance on the use of agentic AI that warns the technology will likely misbehave and amplifies organizations’ existing frailties, and therefore recommend slow and careful adoption of the tech. The agencies delivered that position last Friday in a guide titled Careful adoption of agentic AI services [PDF] that opens with the observation that “Agentic artificial intelligence (AI) systems increasingly operate across critical infrastructure and defense sectors and support mission-critical capabilities,” making it “crucial for defenders to implement security controls to protect national security and critical infrastructure from agentic AI-specific risks.” The thrust of the document is that implementing agentic AI will require use of many components, tools, and external data sources, creating an “interconnected attack surface that malicious actors can exploit.” “Consequently, every individual component in an agentic AI system widens the attack surface, exposing the system to additional avenues of exploitation,” the document warns. To illustrate the risks agentic AI poses, the document offers the example of an AI agent empowered to install software patches that is thoughtlessly given broad write access permissions, with the following unpleasant results: Here’s another nasty agentic mess the document uses as a warning: An organization deploys agentic AI to autonomously manage procurement approvals and vendor communications, and gives the agent access to financial systems, email and contract repositories; This user only considers permissions for the agent when deploying it; Over time, other agents rely on the procurement agent’s outputs and implicitly trust its actions; A malicious actor compromises a low-risk tool integrated into the agent’s workflow and inherits the agent’s over-generous privileges; The attacker uses that privileged access to modify contracts and approve unauthorized payments, and evades detection by creating faked audit logs that don’t trip alerts. Australia’s Signals Directorate and Cyber Security Centre (ASD’s ACSC) contributed to the document, working with the USA’s Cybersecurity and Infrastructure Security Agency (CISA) and National Security Agency (NSA), the Canadian Centre for Cyber Security (Cyber Centre), the New Zealand National Cyber Security Centre (NCSC-NZ) and the United Kingdom National Cyber Security Centre (NCSC-UK). The document contains more scary stories, then lists 23 different risks and over 100 individual best practices to address them. Much of the advice targets developers who deploy AI, but the authors also urge vendors to ensure they test their wares thoroughly and ensure their products “fail-safe by default requiring agents to stop and escalate issues to human reviewers in uncertain scenarios.” The document also urges security practitioners and researchers to spend more time contemplating AI. “Threat intelligence for agentic AI systems is still evolving, which can introduce significant security gaps,” the document warns, because resources like the Open Web Application Security Project and MITRE ATLAS currently focus on LLMs. “As a result, some attack vectors unique to agentic AI may not be fully captured or addressed.” Given the huge to-do list for anyone creating agentic AI, or contemplating its use, the document argues for very cautious adoption. “Organisations should therefore approach adoption with security in mind, recognizing that increased autonomy amplifies the impact of design flaws, misconfigurations and incomplete oversight,” the document concludes. “Deploy agentic AI incrementally, beginning with clearly defined low-risk tasks and continuously assess it against evolving threat models.” “Strong governance, explicit accountability, rigorous monitoring and human oversight are not optional safeguards but essential prerequisites. Until security practices, evaluation methods and standards mature, organisations should assume that agentic AI systems may behave unexpectedly and plan deployments accordingly, prioritizing resilience, reversibility and risk containment over efficiency gains.” ®

Prioritize resilience over productivity, say CISA, NCSC and their friends from Oz, NZ, Canada

Information security agencies from the nations of the Five Eyes security alliance have co-authored guidance on the use of agentic AI that warns the technology will likely misbehave and amplifies organizations’ existing frailties, and therefore recommend slow and careful adoption of the tech.…

Britain's cyber agency is warning that AI-fuelled bug hunting is about to flush out years of buried flaws, leaving defenders scrambling to keep up. In a blog post on Friday, Ollie Whitehouse, CTO of the UK's National Cyber Security Center, said organizations should brace for a looming "patch wave," driven by a backlog of weaknesses now being exposed faster than many teams can realistically fix them. "All organizations have 'technical debt'; a backlog of technical issues – that is both expensive and time-consuming – as a result of prioritising short-term gains over building resilient products," Whitehouse wrote.  "Artificial Intelligence, when used by sufficiently-skilled and knowledgeable individuals, is showing the ability to exploit this technical debt at scale and at pace across the technology ecosystem," he added. The result, according to NCSC, is likely to be a "forced correction" as those weaknesses are uncovered and addressed in bulk. That warning lands just as vendors roll out tools built to do exactly that. Models like Anthropic's Claude Mythos and OpenAI's GPT-5.5-Cyber promise to find and fix bugs before attackers do, but the same capability also lowers the barrier to finding them in the first place. "We are expecting an influx of updates to address vulnerabilities across all severities, and expect a number to be critical," Whitehouse wrote. The cyber agency is urging teams to get ahead of the incoming flood by shrinking their exposed footprint. "All organizations must take steps to identify and minimise their internet-facing (and other externally-exposed) attack surfaces as soon as is possible," Whitehouse said, adding that defenders should "prioritise technologies on your perimeter and then work inwards." Even then, patching alone will not be enough; Whitehouse notes that unsupported or end-of-life systems may need to be replaced altogether. "Prepare to patch quickly, more often, and at scale," is the message from the NCSC. In practice, that means a lot more fixes landing at once, and a lot less time to get them done. ®

Britain's cyber agency says the bill for years of technical shortcuts is coming due, and it's arriving all at once

Britain's cyber agency is warning that AI-fuelled bug hunting is about to flush out years of buried flaws, leaving defenders scrambling to keep up.…