An open, fact-based benchmark
Your personal data is scattered. Who controls it?
Every company you use keeps a copy of your digital life. We grade how well you can see it, get it, and delete it — the same six questions, every company, every fact sourced. Now we also grade the AI agents that act in your name.
Agents at a glance
All 13 agents →| Agent | Where it runs | Agent Safeguardsout of 10 | Policy gradegoverning policy |
|---|---|---|---|
| Underdog | On your device | 7/10 | C6/12Conway Research |
| Claude Cowork | Cloud, per task | 5/10 | B10/12Anthropic |
| Siri AI | On your device | 5/10 | A11/12Apple |
| Grok Bot | Cloud, always on | 0/10 | D4/12Cursor (Anysphere) |
| Instinct | Cloud, always on | 0/10 | D3/12Spear Street Technology |
The three highest and two lowest Agent Safeguards scores.
Company report cards
grade
grade
The Policy grade adds the six scores, out of 12, and gives a letter from A to F.
How we grade — the open methodology
One set of principles. Two rubrics, because companies and agents raise different questions.
What every grade rests on
- Cell scores rest only on a company's own published statements.
- Regulator decisions feed a separate enforcement record, never a cell.
- Press and advocacy filings are leads, not sources.
- The same public criteria for every company. Every fact sourced and dated.
- Companies can climb. Every change is logged.
Policy grade, 0–12
Six cells: Access, Portability, Friction, Deletion, Sells / shares, Trains AI. Letter grade A–F. Graded per account and per region. Full review every 12 months, with a daily watch of every cited page.
Under Portability, each company page also states whether apps can connect to your data with your permission. That is a fact, not a score.
Read the company rubric →Agent Safeguards, 0–10
Five cells: Logins, Capture, Memory, Where it runs, Exit. Shown as a number out of 10, with no label. Full review every 90 days, every 30 in beta, and within 7 days of a launch, rename or shutdown.
Each agent also carries the Policy grade of the policy that governs it. Self-hosted tools get a context card, not a grade.
Read the agent rubric →Every company answers the same six questions — Access, Portability, Friction, Deletion, Sells/shares, Trains AI — each scored Best (2) / Moderate (1) / Restrictive (0). The six add to a total out of 12 on a flat, published scale (A 11–12 · B 9–10 · C 6–8 · D 3–5 · F 0–2), and a worst score on Sells/shares or Trains AI caps the grade at C. A verified regulatory action then deducts points off the total (§). Every cell rests on a sourced, dated fact, assessed per jurisdiction.
9 REGIONS VERIFIED · MORE ADDED ON DEMAND The US, EU, UK, Brazil, India, Indonesia, Japan, Canada and Australia columns are source-verified against each company's own pages plus the local legal floor, with live links and a review date. We do not grade a single "Rest of world" column — it spans 170-plus jurisdictions, so we add new countries one at a time as we can verify them. China is shown but not scored, with an explanation. Hover any cell for the deciding fact. Contributed requests sharpen our confidence and flag cells for re-review, but won't on their own move a score.
Can you see all of it?
View what the company holds on you — every major category, clearly, self-serve, free.
Can you get it out in a usable form?
Export it in a structured, machine-readable form — not a scrape or PDF dump (the data-portability right).
How hard and how slow is it?
How fast and painless the process is — and whether dark patterns block the way (the speed nuance).
Can you actually delete it?
Actually delete it — not just deactivate — with clear scope and timeline.
Who else gets it?
Whether it's sold, shared, or brokered — and whether the opt-out works.
Do they train their AI on you?
Whether it trains AI on your data — and whether the opt-out is real and findable.
Under Portability, each company page also states whether apps can connect to your data with your permission: Connects to apps — Yes, Partly, On your device only, or No. That is a fact, not a score.
A verified regulatory action — a real, dated, regulator- or court-confirmed order, settlement, or fine — deducts points off the /12, on top of the six scored questions. Severity 1 deducts 2 points and covers verified actions such as opt-out failures, deceptive ad-targeting, or disclosure lapses; severity 2 deducts 4 points and is reserved for the most serious records — unlawful large-scale data transfer, children's data, a breach causing demonstrable harm, or documented repeat offenses. Severity tracks the nature of the violation, not the size of the fine. The deduction sits at full weight for the first ~5 years, then ramps down to zero by ~8 years if no further verified action resets the clock — the record stays visible even after it stops moving the grade. We state only the verifiable fact (the action, its date, its source), never an inference of reform. Companies with no verified action carry no penalty. Each penalised company shows the full arithmetic — practices total, deduction, and final grade — with the dated evidence.
A grade is a documented assessment against a public rubric, as of the reviewed date — not a security audit, legal ruling, or guarantee of any outcome.
Your data, in your hands
Your data never passes through us — anything you retrieve goes straight to your own device. Start with a company above, or open your own list.