AI that helps your team think. Never AI that labels a child.
Threat assessment carries real consequences for students. We built VeriThreat's AI to reduce the clerical burden on teams and surface what's missing, and we deliberately left out the things AI shouldn't do.
There is no profile of a school attacker. AI can't find one either.
The Secret Service found that attackers varied in age, gender, race, grade, academic performance, and social characteristics, and that profiles and checklists don't yield accurate or useful results. A model trained to rank students would learn to flag the students it sees most often in disciplinary data, not the ones who pose a risk.
That's why VeriThreat has no risk scores, no watch lists, and no “students to watch” dashboard. The AI only ever looks at behavior that a person reported, and it hands its work back to that team.
- Suggest queue urgency for new reports
- Organize records under NTAC's 13 themes
- Cite the source of every finding
- Name information gaps and people to consult
- Suggest supportive management options
- Point out conflicting or weak evidence
- Scores or ranks the student body
- Sees race, religion, disability, or income
- Lowers an emergency a person flagged
- Sets the team's level of concern
- Contacts students, parents, or police
- Runs without a human request or report
Four guardrails on every AI request.
- 01
Only reported concerns
AI runs on a report someone filed, or on a case your team opened. It never scans or scores the general student body.
- 02
Minimized and pseudonymized
Students become S1, S2, and so on. Known names are removed from free text. IDs, birth dates, and school names are never sent.
- 03
Grounded in the record
Every finding must cite the report, note, or attachment it came from. Citations that don't match a real record are removed.
- 04
Reviewed by people
Output is labeled advisory, shown separately from the team's decision, and only a BTAM lead can record a level of concern.
“Unknown” is a feature.
Early in a case, most themes are unknown, and saying so is the most useful thing an assistant can do. VeriThreat's analysis marks each theme as concern indicated, no concern indicated, or unknown, explains why, and tells your team where to look next.
- Suggests “insufficient information” rather than guessing a level of concern
- Separates what was observed from what was reported secondhand
- Treats the text in reports as evidence to assess, never as instructions to follow
- Writes plainly, for a file that parents or courts may one day read
Your district stays in charge.
Off switch
On-demand analysis
Fail-safe defaults
Human sign-off
We recommend districts evaluate any AI-assisted triage on de-identified historical cases, including checking for disparities across student groups, before relying on it.
Every attacker in the Secret Service's study showed concerning behavior first.
Make sure the people who notice it have somewhere to go, and your team has what it needs to act.