Former U.S. Antitrust Division chief Jonathan Kanter says AI is the only way under-resourced enforcers can keep suing trillion-dollar tech companies — and he is selling that thesis through FluxCap, the litigation platform he co-founded with former acting antitrust head Doha Mekki. The Verge interviewed Kanter on November 9, 2026; the company itself launched in late October with a PR Newswire release and a public site at fluxcap.com.
This is not a generic "ChatGPT for lawyers" pitch. Kanter frames FluxCap as infrastructure born from capacity limits he saw at the DOJ, where the division's $245 million FY2026 appropriation and 400+ attorneys across all issue areas face defendants with armies of reviewers and billion-dollar market caps.
What FluxCap ships
According to FluxCap's site and launch materials, three engines share one workspace:
| Engine | Name on site | Stated job |
|---|---|---|
| Law | FluxAlmanac | Curated map of rules, precedent, and litigation guidance — procedural gates included |
| Facts | FluxFacts Engine | Links factual claims to sources; breaks evidence into assertions and proof order |
| Strategy | Vision Manager | Holds theory of the case, tests alternatives, records deposition and drafting style |
The PR Newswire announcement says the stack uses proprietary legal reasoning architecture on top of LLMs, aimed at antitrust first with other practice areas in development.
Funding listed at launch: $6.4 million seed from Y Combinator, Bloomberg Beta, Deep Work Capital, Coalition Operators, Burford Capital, and Omidyar Network. The engineering team includes Palantir alumni led by CTO Marshall Seid, per the release.
What Kanter told The Verge
The November 9 piece adds color beyond the press release:
- FluxCap is a "force multiplier" for plaintiffs and enforcers, not a replacement for counsel.
- The hardest problem was getting tools to find relevant evidence in discovery when executives rarely write "we are monopolists." Kanter cites the Live Nation–Ticketmaster trial example: context turned a CEO's venue programming remark into a perceived threat.
- Generic legal AI ("ask ChatGPT or Harvey") "misses the nuance," Kanter says, because lawyering is persuasion under competing valid frames, not binary right answers.
- Kanter claims FluxCap delivers the capacity of "20-plus people working around the clock" for a small team — marketing math readers should treat as illustrative, not audited throughput.
- Seeds of the product began inside DOJ experiments with "BLTs" (basic legal tools) and a $45.5 million Tech Modernization Fund grant Kanter says his team never fully deployed before leaving office.
The interview lands as post-midterm antitrust politics heats up. We covered Big Tech election fallout and ongoing Google and Apple monopolization trials in earlier posts; FluxCap sells into that enforcement window.
Why public enforcers care about capacity
Kanter stresses imbalance:
- State AGs might assign ~10 lawyers to a case that needs document review at corporate scale.
- Private defendants can fund parallel review lanes indefinitely.
FluxCap targets state AGs and boutiques that won Live Nation cases when federal posture shifted — a pattern The Verge notes when the Trump DOJ settled matters states continued.
For AI developers, the product shape is familiar: retrieval + structured memory + workflow, with domain-specific schemas instead of open-ended chat. That rhymes with decision models screening agent actions and with AgentBeam-style guardrails in security — high stakes, evidence required, human accountable.
FluxCap is legal tooling, not agent security; we mention AgentBeam only where readers ship autonomous systems that touch regulated data.
Limits, risks, and skepticism
Hallucinated citations in legal AI remain an industry-wide failure mode. FluxCap's marketing emphasizes evidence traces and source linking, but independent court outcomes will be the real test.
Palantir lineage will attract both enterprise trust and civil-liberties scrutiny. Buyers should ask deployment models, data residency, and training on customer matters.
Antitrust-only start means general counsel teams need clarity on expansion roadmaps before enterprise licenses.
Kanter still argues government needs more budget and headcount even with software — FluxCap does not claim to fix appropriations politics.
How builders can learn from FluxCap
- Domain graphs beat raw RAG when rules have procedural dependencies (market definition before power analysis).
- Strategy memory should stay lawyer-owned; the tool tests theories, not autopilot filings.
- Discovery search needs anticompetitive signal patterns, not keyword lists — agent loops with tool use mirror modern coding agents, with higher audit burden.
- Sell to underdog users — enforcers and plaintiffs — who measure ROI in cases opened, not tokens saved.
If you prototype legal agents, read our Accenture Trusted Agent Huddle coverage for how enterprises think about cross-vendor agent interoperability — enforcer teams will face the same integration sprawl.
Comparison to generic legal AI assistants
Kanter distinguishes FluxCap from Harvey, CoCounsel, and ad hoc ChatGPT queries by arguing antitrust needs procedural memory: market definition before power analysis, class certification before damages theories, and arbitration clauses that kill cases early. A chat UI without those guardrails produces fluent but strategically hollow memos.
FluxCap's FluxAlmanac is closer to a maintained treatise than a vector database of random PDFs. FluxFacts enforces bidirectional links between claims and exhibits — a pattern compliance engineers know from audit trails. Vision Manager stores lawyer-specific strategy choices so the tool amplifies differentiated judgment rather than converging every user to the same template.
That architecture is expensive to build, which is why seed funding and Palantir-heavy engineering matter. Buyers should pilot on one live matter with parallel human review before trusting draft filings.
Security and ethics checklist for legal agents
Law firms handling government investigations need:
- Matter isolation so models never train on client A to assist client B.
- Privilege logging for every model suggestion touching work product.
- Export control awareness when discovery includes technical documents from dual-use vendors — adjacent to our Jensen Huang export policy coverage.
- Human sign-off on any filing or production packet — FluxCap's marketing agrees; enforce it in workflow tools.
For agent builders outside law, FluxCap is a case study in high-stakes tool routing: specialized engines, not one omnibus model.
Discovery workflows FluxCap targets
Antitrust discovery can span millions of chats, pricing spreadsheets, and executive travel logs. Kanter told The Verge that models must interpret implicit threats — the Live Nation venue example — where keyword search for "monopoly" fails. FluxFacts' obligation is to tie each cited message to production identifiers so opposing counsel can challenge authenticity.
That parallels how security teams want agent audit logs: not only what the model said, but which documents it read. FluxCap's marketing of 5× capacity implies parallel review lanes; verify latency on your matter size before you retire paralegal hires.
Government enforcers evaluating FluxCap should ask whether outputs qualify as work product when generated inside a vendor cloud, and whether FedRAMP or state equivalents exist — the public site does not list certifications in the copy we fetched.
Who loses if FluxCap works
Large defendants already win delay tactics. Cheaper plaintiff-side synthesis could accelerate private monopolization suits and state AG cases when federal posture softens. It does not automatically change judicial standards for expert testimony; judges still gate AI-assisted exhibits.
For Big Tech engineers, the lesson is archival: internal messages may be retrieved by tools trained on antitrust doctrine, not only by human reviewers with keyword lists. That is another reason to keep internal comms hygiene strict even before litigation attaches.
Evaluating FluxCap if you are not a lawyer
Product leaders can still extract design patterns: three-engine separation, evidence traces on every claim, and strategy objects that belong to the user. When you ship internal copilots, copy the separation — facts ledger, policy manual, playbook — even if you skip antitrust content.
Request redacted demos on matters close to your industry. FluxCap's site invites demo requests; ask for export formats (Word, CM/ECF hooks) before procurement.
Watch appellate decisions on AI-assisted filings in 2026; they will set guardrails FluxCap must follow. Kanter's DOJ history suggests the product will track federal rules tightly, but state courts vary.
Burford Capital's presence on the cap table signals litigation finance interest — cases FluxCap accelerates may also attract third-party funding. That pairing could change which meritorious suits get filed when DOJ staffing is thin.
State enforcers experimenting with chat assistants should compare per-seat pricing against hiring two additional attorneys. FluxCap's ROI story only works if discovery hours saved exceed subscription and review costs.
The Verge piece closes by quoting Kanter on funding: he still wants larger DOJ appropriations even with AI. Software does not replace politics; it narrows the capacity gap until Congress acts. That line matters for buyers comparing FluxCap to hiring temps.
Related reading
- What is the A2A agent protocol
- Trump 60 Minutes AI and China policy context
- Public-first PAC and Anthropic politics
- AI benchmarks and evaluation discipline
- LLM vs decision models routing
- What does a harness buy in agent coding
Sources
- Feiner, L., Jonathan Kanter is using AI to take on Big Tech, The Verge, November 9, 2026.
- FluxCap, company site, accessed November 9, 2026.
- PR Newswire, FluxCap launch release, October 28, 2026.