Threat reportThreat IntelligenceTL-2026-3659
Hyperlocal fake news website networks targeting AI chatbots to influence election-related responses
Hyperlocal fake news website networks targeting AI chatbots (TL-2026-3659) is a medium-severity tracked intrusion set, first published 2026-10-15. It has no confirmed attribution, affects Perplexity Perplexity AI search/answer engine, maps to 6 MITRE ATT&CK / ATLAS techniques (AML.T0020, T1565, T1583.001), and is covered by 9 detection rules and 5 indicators of compromise.
- Severity
- MEDIUMAssessed severity
- CVEs
- 0None referenced
- Techniques
- 6MITRE ATT&CK / ATLAS
- Actors
- 0Not attributed
- Detection rules
- 9SPL · KQL · Sigma
- IOCs
- 5Indicators of compromise
Key facts for TL-2026-3659
- Threat ID
- TL-2026-3659
- Severity
- MEDIUM
- Status
- ACTIVE
- Category
- THREAT_INTEL
- First published
- Last reviewed
- Attribution confidence
- LOW
- Motivation
- UNKNOWN
- Target sectors
- government administration, elections, news - media, technology
- Target regions
- North America
- Detection rules
- 9
- Indicators of compromise
- 5
How Hyperlocal fake news website networks targeting AI chatbots works
VisionHeight reports networks of hyperlocal fake news websites across 14 US states, configured to allow AI crawlers and filled with repetitive, AI-optimized content favoring political candidates, apparently intended to shape AI chatbot answers. Perplexity was reported to have cited one of the sites (Tarheel State Report) as a source. VisionHeight does not attribute the activity to a specific actor.
Per CyberScoop's reporting of research by Israeli threat intelligence firm VisionHeight, an influence operation is building networks of fake hyperlocal news sites whose primary audience is AI crawlers rather than human readers. VisionHeight initially identified 37 sites across 6 states and later expanded the finding to 460+ pages across 14 states: Arizona, Colorado, Georgia, Indiana, Iowa, Kansas, Michigan, Mississippi, Nevada, North Carolina, Pennsylvania, Texas, Virginia and Wisconsin. Networks typically comprise 5-6 sites per state, with state-flavored names such as Wolverine State Report, Peach State Bulletin and Tarheel State Report.
Infrastructure was stood up in bulk: all 37 initial sites were registered in a single evening, 36 of them within an 8-minute window. Roughly 100,000 words of candidate-related content were published on September 30 using identical templates. The text consists of repetitive candidate-preference statements (for example, 'Candidate X is the stronger choice' on housing, taxes and other issues) with slight linguistic variation, formatted for machine extraction rather than readability. Targeted contests include House seats, Secretary of State races, judicial appointments and local offices. The sites' robots.txt files explicitly permit scraping by major US AI companies, in contrast to most legitimate media sites, which block AI crawlers.
The intended effect is to poison the retrieval/search layer (and potentially future training data) of AI chatbots so that answers about local candidates reflect the planted narrative. Perplexity was reported to have linked to Tarheel State Report as a primary source. AI generation is suspected in both infrastructure and content creation. VisionHeight does not attribute the activity to any actor; CyberScoop notes that domestic or foreign actors could replicate the model at scale within days. This is a technique-level (TTP) record: the source provides no domains, IPs, hashes, malware or CVEs, so the IOC set is limited to named entities and behavioral indicators. Severity is an analyst assessment, not stated in the source.
This pattern is consistent with previously documented 'LLM grooming' (NewsGuard's Pravda-network research) and AI-generated pink-slime local news, which are included as background context only; no link between those operations and this activity is asserted.
MITRE ATT&CK / ATLAS techniques used in TL-2026-3659
persistence
AML.T0020 Training Data Poisoning
Impact
Resource Development
T1583.001 Acquire Infrastructure: Domains; T1588.007 Obtain Capabilities: Artificial Intelligence; T1608 Stage Capabilities
Defense Evasion
Affected products and versions in Hyperlocal fake news website networks targeting AI chatbots
- Perplexity — Perplexity AI search/answer engine
Vulnerable versions: Reported to have cited Tarheel State Report as a source - Multiple — AI chatbots and AI search tools that crawl the open web (sites permit major US AI company crawlers)
Vulnerable versions: Not specified in source
Remediation for Hyperlocal fake news website networks targeting AI chatbots
Immediate actions
- Review AI/RAG and search-assistant source allowlists and de-prioritize or block newly registered, template-heavy hyperlocal news domains
- Flag clusters of domains registered within minutes of each other with identical templates and permissive AI-crawler robots.txt files
- Add human review for election-related answers that cite single-source or low-reputation local news sites
Workarounds
- Election officials and campaigns can test major chatbots for local-race queries and report poisoned citations to the vendors
Longer-term hardening
- Implement source-reputation and domain-age scoring in retrieval pipelines for chatbots and AI search
- Detect near-duplicate content with slight lexical variation across domains as a coordination signal
- Monitor candidate and race names in AI answers for sudden shifts tied to new sources
Timeline of Hyperlocal fake news website networks targeting AI chatbots
- Background context: NewsGuard-based reporting finds partisan/AI-generated 'pink slime' local news sites outnumber real local newspaper sites in the US (not linked to this activity).
- Background context: OECD.AI records an incident on AI-driven fake news sites eroding trust in US elections; prior art for AI-generated fake local news (not linked to this activity).
- Background context: NewsGuard reports the Pravda network flooding the web to influence chatbot outputs ('LLM grooming'); prior art for this TTP, not linked to this activity.
- Background context: Yale ISPS publishes a study finding people often trust fake local news sites more than real ones.
- Roughly 100,000 words of candidate-related content published across the hyperlocal sites using identical templates (per VisionHeight via CyberScoop).
- CyberScoop reports VisionHeight's findings: 37+ sites initially, expanded to 460+ pages across 14 states; Perplexity reported to have cited Tarheel State Report.
- As of CyberScoop's report, Perplexity was confirmed to have cited Tarheel State Report as the first source for a local-race query; Google Search is also named as an AI search surface of concern.
Sources cited for Hyperlocal fake news website networks targeting AI chatbots
- The next election influence campaign will target your AI chatbot (Derek B. Johnson, CyberScoop)
- Context: Russian disinformation 'infects' AI chatbots, researchers warn (NewsGuard Pravda network study coverage)
- Context: NewsGuard Special Report - A well-funded Moscow-based global 'news' network (Pravda network)
- Context: Russian fake news network floods Western AI chatbots with millions of propaganda articles (The Decoder)
- Context: Study - People often trust fake local news sites more than real ones (Yale ISPS)
- Context: Phoney news portals surpass US newspaper sites, researchers say (Malay Mail / NewsGuard)
- Context: AI-Driven Fake News Sites Erode Trust in US Elections (OECD.AI incident)
Detection coverage for TL-2026-3659
As of 2026-10-15, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-3659 across Splunk SPL, Microsoft KQL and Sigma, covering 5 indicator(s) of compromise. The whole corpus is readable without an account; a free account unlocks full detection query text in Splunk SPL, Microsoft KQL and Sigma; paid tiers add raw indicator values, correlation and the MCP server. Threadlinqs MCP server · View plans.