What Is Gideon AI Threat Detection? A Plain-English Guide for 2026

The Gideon Pitch: What's Claimed and What's Verified

On June 2, 2025, Aaron Cohen appeared on Fox News' Hannity and told Sean Hannity, "I have just built, Sean, the first AI-powered threat detection system for law enforcement that scans the open web for radicalization and tactical planning." That television pitch is, to date, nearly everything the public knows about Gideon AI Threat Detection — and much of it remains unverified.

According to Mediaite's report on the Fox News appearance, Cohen — an Israeli special operations veteran — described Gideon as a system that scans the open web and social media for "pre-attack signals" including radicalization, manifesto writing, tactical planning, target casing, and group coordination, then pushes real-time alerts to police and school threat units. He claimed his engineering team includes former Palantir engineers and urged the Trump administration to "partner with me" and scale the system nationally.

But here's the critical gap readers need to understand before treating Gideon as a real 2026 product. A September 2025 critical report states Gideon "doesn't appear to even exist yet," and that Cohen was reportedly seeking seed funding to build a minimum viable product. Cohen's own website, per that report, solicits donations to build Gideon through his online store — which sells holsters, books, coffee, and apparel. He claimed during a September 2025 Fox News appearance that Gideon was "launching next week," though no funding or launch has been confirmed.

The claimed-vs-verified ledger looks like this:

• Claimed: 24/7 scanning of social media, gaming chats, Discord, Reddit, and fringe blogs, marketed as "a 24/7 detective that never sleeps," per a critical advocacy analysis.
• Claimed: a dozen agencies "already on board," including a major Northeast police department — though no agencies are named.
• Claimed: "100 percent" confidence Gideon could have prevented the Minneapolis school attack.
• Verified: nothing. No pricing, no technical specifications, no named customers, no independent performance data exists in any public source.

That "100 percent" claim deserves special scrutiny. Vendor guidance on AI threat detection is blunt: "If a company promises 100% detection, walk away. No system can deliver that," according to development industry analysis. Even peer-reviewed research concludes AI threat detection "is not a standalone solution" and must be transparent and explainable.

One disambiguation note: don't confuse Gideon AI with Scanary, an Israeli AI weapons-scanning startup led by Dr. Gideon Levita — a person named Gideon, not the product.

Cohen's story is a case study in the "verify before you trust" principle. It's the same principle behind SkillzBench, which is building free security scanning and benchmarking for AI agent skills — because published research found roughly 26% of agent skills carry at least one vulnerability. Claims are cheap; verification is the product.

CTA: Agent skills can automate your threat-intel and security workflows — but only ones that pass verification. Browse and test free skills on SkillzBench. Skills, proven.

How It's Supposed to Work — and Where It Would Be Deployed

According to Cohen's own pitch on Fox News' Hannity in June 2025, Gideon scans the open web and social media for "pre-attack signals" — radicalization, tactical planning, manifesto writing, target casing, and group coordination. One critical report describes the claimed scope even more broadly: social media, gaming chats, Discord, Reddit, fringe blogs, and purchase histories run through a proprietary model, marketed as "a 24/7 detective" that never sleeps. The system would then push real-time alerts to law enforcement or school threat units so they could act before an attack happens.

The intended deployment picture is fairly consistent across sources, even if the product itself is not:

• Local law enforcement agencies, described as the primary end user.
• School threat units and safety teams, with the tool explicitly marketed for school shooting and mass violence prevention.
• National scaling, which Cohen sought by asking the Trump administration directly to "partner with me and let's scale this."

Cohen claimed a dozen agencies were already on board ahead of launch, including a major Northeast police department — though no agencies are named. He also claimed his engineering team includes former Palantir engineers, describing Gideon as "Palantir-quality engineering" based on "the Israeli model."

Here's the crucial caveat: whether Gideon actually works this way — or exists at all — is unverified. A September 2025 report concluded that Gideon "doesn't appear to even exist yet" and that Cohen is reportedly seeking seed funding to build a minimum viable product. His website, per that same report, offers donations toward building Gideon alongside holsters, books, and a coffee brand. All feature claims trace back to Cohen's own promotional statements, with no independent technical documentation or third-party evaluation.

On pricing, the honest answer is that no public source provides any. No pricing, deployment costs, or technical specifications exist anywhere in available coverage — which itself is telling for a product pitched as ready for national rollout.

One naming note to prevent confusion: Gideon AI is not Scanary. Scanary is a separate Israeli startup, founded in 2024 in Tel Aviv and led by Dr. Gideon Levita — a founder named Gideon, not a product. Scanary makes a radar-based AI weapons scanner that screens up to 25,000 people per hour in under two seconds per scan, and has raised $3.5 million. The two entities are unrelated, but the shared name makes search results easy to conflate.

For anyone evaluating AI security tools generally, the same verification principle applies: trust claims only when they're independently testable. That's the philosophy behind SkillzBench, which is building free security scanning and benchmarking for AI agent skills — published research found roughly 26% of agent skills carry at least one vulnerability, so verifying before deploying matters whether you're assessing a surveillance pitch or an automation workflow.

The Civil-Liberties Backlash and the Red Flags Every Buyer Should Know

The backlash to Gideon-style surveillance isn't abstract. Privacy advocates argue that scanning the open web and social media around the clock — social posts, gaming chats, Discord, Reddit, fringe blogs — amounts to mass surveillance without warrants or individualized suspicion. Critics frame it as "pre-crime": flagging people for what they might do, not what they've done.

That framing raises direct constitutional questions. Monitoring people for "martyrdom language" or "target casing" before any crime has occurred implicates the First Amendment, and civil rights groups argue "a computer's judgment" cannot serve as probable cause or reasonable suspicion under the Fourth Amendment. Jeramie Scott, senior counsel at the Electronic Privacy Information Center, describes comparable tools as part of an "unchecked surveillance ecosystem" lacking meaningful transparency, oversight, or accountability — a description critics say maps well onto Gideon.

The chilling effects are measurable, not hypothetical:

• After the Snowden revelations, traffic to Wikipedia articles on sensitive topics like terrorism dropped by roughly 30 percent.
• 1 in 5 U.S. households reported curbing their internet activity over worries about government data collection.
• The Flock automated license-plate reader system has already produced wrongful arrests and at least one case of an officer stalking someone.

There's also the accuracy problem — and here Gideon's story offers a lesson that applies to every AI threat detection product, security or otherwise. Cohen claimed "100 percent" confidence that Gideon could have prevented the Minneapolis school attack, yet no independent performance data exists for the system. Meanwhile, peer-reviewed research on AI threat detection is blunt: black-box opacity, false positives, and the need for human oversight are persistent, unsolved limitations. As one industry guide puts it, "If a company promises 100% detection, walk away. No system can deliver that."

Predictive systems built on digital footprints carry documented risks of false positives, racial profiling, and political targeting, according to privacy advocates, legal scholars, and former intelligence professionals. A model that flags "tactical planning" in a gaming chat can't explain why it flagged it — and a person's life can change on the strength of an unexplainable alert.

The takeaway for buyers evaluating any AI security tool: demand evidence, not confidence. Ask for independent validation, published accuracy rates, and explainability before trusting a vendor's claims. The same "verify before you trust" principle applies to the AI tools businesses deploy daily — published research found roughly 26% of AI agent skills carry at least one vulnerability, which is why SkillzBench is building free security scanning, benchmarking, and a free MCP connector so teams can verify what they install before it touches their systems. Confidence is marketing. Evidence is security.

AI Threat Detection That Actually Ships: Lessons for Your Own Stack

Gideon's story is a masterclass in what happens when AI security claims outrun evidence — and it's a lesson that applies directly to your own stack. If a vendor claims "100 percent" detection, walk away, because no system can deliver that, as one industry guide on building AI threat detection systems puts it bluntly.

So what does verified AI threat detection actually look like today? It starts with anomaly detection: unlike traditional rule-based tools, AI-powered systems can catch previously unseen attacks and continuously adapt to evolving techniques, according to Oligo Security's technical explainer. Common, proven use cases include identity visibility, misconfiguration detection, brute-force and MFA-bombing alerts, automated remediation, and SIEM/SOAR integration.

The payoff is measurable. IBM's 2025 Cost of a Data Breach Report found that heavy AI and automation users saved roughly $1.9 million per breach and cut breach timelines by about 80 days — against a global average breach cost of $4.44 million and 241 days to find and contain an incident. But the academic consensus is equally clear: AI "is not a standalone solution," and most systems remain black boxes, so human analysts must own judgment, investigation, and strategy.

When evaluating any AI security vendor, demand:

• Independent evidence and benchmarks, not founder claims made on TV
• Transparency about how decisions are made — no black-box opacity
• Named customers and verified deployments you can check
• Human oversight built into the workflow, not bolted on

The same "verify before you trust" principle applies to anyone deploying AI agents. Published research found that roughly 26% of AI agent skills carry at least one vulnerability — meaning nearly one in three instruction packages you might wire into Claude Code, Cursor, or another agent could introduce risk. That's why SkillzBench is building a free security-scanning, benchmarking, and MCP-connector layer: pin every skill version, scan it, and test it in a sandbox before it ever touches your agent.

Gideon's pitch sounded impressive; the verification was missing. Whether you're buying a threat detection platform or installing an agent skill, insist on evidence first.