Industries / Finance and business services / Security guard services
Private investigation
Private investigation is fact-finding done for pay by investigators outside the police, mostly for lawyers, insurers and businesses: searching records, checking companies, looking into doubtful claims and writing up the findings in reports that may be used in court.
1833
In Paris, Eugène François Vidocq, a former convict who had twice run the police's plainclothes Sûreté unit, founded Le bureau des renseignements ("Office of Information"). It began with eleven detectives, two clerks and a secretary, and worked for businesspeople and private citizens against fraudsters and bankruptcy artists. Wikipedia and a 2017 Salon article both say it is considered the first private detective agency.
31
Milestones achieved across 5 domains.
7
To watch for: announced, not yet confirmed outside.
10
Open deliverables. Nobody has shipped them.
In one minute
Where it started
Private investigation became a business in Paris in 1833, when Vidocq opened an agency that took cases for businesspeople and private citizens. In Chicago, Allan Pinkerton set up a detective agency that chased railway thieves, around 1850 by the US National Park Service's account. In New York the Mercantile Agency, which grew into Dun & Bradstreet, had begun reporting on merchants' credit in 1841, and California began licensing private detectives in 1915.
Where it stands
The US Bureau of Labor Statistics counted about 39,500 private detectives and investigators in 2025, with a median wage of $51,220. Retail trade employed 34% of them, and finance and insurance 11%. 45 of the 51 US jurisdictions license investigators under state law, by the count of the United States Council for Law & Investigation (USCLI), while the United Kingdom never brought in the license it promised in 2013. Several of the main legal limits on the work followed cases in which investigators got personal records by deception, or for clients who misused them.
What is open
Most of the open jobs are tests that nobody has run on the tools investigators are starting to use. AI research and report-writing tools are sold on their makers' own figures, and the only controlled trial of AI report drafting found in this research, run with police, measured speed and not accuracy. Each job below uses fictional case files, company records or the text of laws, so no real person is investigated.
Domains
AchievedWatch forOpen
Domain
Progress
Last achieved
Watch for
Open
Public records research
Public records research is finding and reading the records that governments and courts keep, such as court files, company filings and vehicle records, and the commercial databases that resell them.
6 · 1 · 2
Last achievedDriver data sales counted2025
Watch forAgentic records research
2
openInsurance and fraud cases
Insurance and fraud work is checking claims that may be invented or inflated, usually for an insurer's special investigations unit (SIU), by reviewing the claim file, taking statements and checking documents.
5 · 2 · 2
Last achievedClaim photos checked across insurers2023
Watch forDeepfake alerts in claims
2
openChecks on businesses
Checks on businesses, called due diligence, are finding out what a company is, who owns it, and what record it has of fraud or lawsuits, before a client invests in it, buys it or trades with it.
6 · 2 · 2
Last achievedDue diligence as data gathering2023
Watch forOwnership traced by AI agents
2
openEvidence and reports
Evidence and reports is collecting photos, documents, digital files and interviews so that they can be shown to be genuine, and writing the report that a client or a court will rely on.
7 · 1 · 2
Last achievedDeepfake exhibit ends a case2025
Watch forA rule for machine evidence
2
openLicensing and law
Licensing and law covers who may work as a private investigator, what they must show to get a license, and which laws limit how they gather information.
7 · 1 · 2
Last achievedA privacy code for investigators2024
Watch forA code as UK rulebook
2
open
Advances to build with
3 research notes

Claude Fable 5.2
Three working devices
Anthropic says its AI model Claude Fable 5.2 designed three prototype devices that worked without electrical modifications, with no outside test reported.

Muse Ember
The jobs agents repeat
Meta's Muse Ember matched Claude Haiku 5.5 on an outside firm's tests at a third of the cost per task, though both list the same prices for text in and out.

Cursor Agent SDK
Findings that cite notes
Cursor released a kit for building AI agents, and its example, a code reviewer that cites the note behind each finding, is known only from Cursor's account.
Open in private investigation
3 of 10 open deliverables
Open · Public records research
A test town for agents
A fictional county, served from a local web server, with 20,000 public records about 2,000 fictional businesses: court dockets, liens, property deeds, company filings and business licenses. It has planted traps, such as near-identical company names, a record filed in the neighbouring county, a dissolved company later reinstated, and an index that contradicts the document it points to. 300 research tasks have known answers, each tied to the record that proves it, to score AI research agents.
Built with what exists
Claude Fable 5.2 to write the fictional records and their contradictions (see the Note "Three working devices"); a small web app that copies the layout of real county record portals; Browser Use 1.0, an open-source agent that writes Python to drive Chrome, so that agents use the portals as a person would (see the Note "Deleting 80% to start over").
Open · Insurance and fraud cases
File review for SIU work
200 fictional claim files of the kind an SIU investigator reviews: claim form, recorded statement transcript, police report, repair estimate, invoices, and photos with their metadata. Some carry planted inconsistencies, such as a loss date that does not match a photo's metadata, a statement that contradicts the police report, or an invoice for parts not in the estimate, all listed in an answer key. AI tools and experienced investigators are each asked to list every inconsistency and the page it is on.
Built with what exists
Claude Fable 5.2 to write the files (see the Note "Three working devices"); ElevenLabs Scribe v3 to transcribe statements recorded by volunteer actors reading scripts (see the Note "Fewer word errors"); a Claude-based reviewer of the kind Enveda describes for drug-making records, which raised issues people had missed (see the Note "Seven more flagged").
Open · Checks on businesses
Parent companies across borders
A test of whether AI agents can find a company's direct and ultimate parent when the chain crosses borders, using only public registers and filings. It stops at companies and never names people. The answer key is GLEIF's Level 2 data, which covers parent companies, not people. It uses 500 companies in at least 20 countries, each with a parent in another country.
Built with what exists
GLEIF's free API and daily relationship files; the Companies House API; the national registers on GLEIF's Registration Authorities List; Browser Use 1.0 for registers that have no API (see the Note "Deleting 80% to start over"); Meta's Muse Ember, a low-cost model, for the many repeated lookups (see the Note "The jobs agents repeat").
Where it sits
Broader