iDharma

Name: iDharma
Address: Greenwood, IN, US
Telephone: +13175055697

Independent AI audits for enterprises - proving your AI is accurate, fair, secure & compliant. iDharma provides independent AI audits for enterprises.

We verify that your AI systems are accurate, fair, secure, and compliant - then deliver a clear, evidence-based report and a practical fix path. As an independent third party, we assess AI models and AI-powered workflows against real-world risk, regulatory, and governance standards - so leaders can trust their AI and prove it to customers, regulators, and boards. Know your AI is accurate, fair, secure & compliant. Learn more at idharma.us

09/22/2026

Three supervisors rewrote model risk guidance. Not one wrote a new rule for AI.
They didn't need to — an AI model is a model. SR 26-2, SS1/23 and OSFI E-23 reach the same question from three different directions: who validated this, independently of whoever built it? Brief #006 puts all three side by side, dated and sourced. Including the one artefact we ask for before we look at the model at all.

If your AI models sit outside your model inventory, that isn't an AI gap. It's a model risk gap — and it has been findable for years.

Archive link is in the first comment — subscribe there if you want #007 next Tuesday.

09/18/2026

"We think we're compliant" is not evidence. It's a belief about your own homework — and it's the sentence that appears right before a regulator asks for the test result.

Evidence has four parts: a named obligation, a test that was run, a result with a date on it, and a person whose name is on that result. Miss one and it isn't evidence. Miss all four and it's a feeling.

Article 50 transparency duties have applied since 2 August 2026; stand-alone Annex III high-risk obligations apply from 2 December 2027. The gap between them is when evidence is cheap to build.

Comment AUDIT and we'll send you the teardown.

09/16/2026

Same model, two products: one is high-risk under the EU AI Act and one isn't. The architecture is identical — the purpose is what differs.

Comment AUDIT.

09/15/2026

Brief #005
Two dates decide your EU AI Act position, and most people have only heard one. Annex III high-risk duties now apply from 2 December 2027 — but Article 50 transparency has applied since 2 August 2026.
Brief #005 walks the high-risk list: the eight areas, what the Digital Omnibus actually changed when it entered into force on 27 July 2026, and what is already enforceable today.
Subscribe · archive in the first comment.

09/11/2026

The finding takes one line to write and most employers can't dispute it: the published bias audit is more than twelve months old. Nothing was hidden - the date simply went stale.
Here's how it appears in a report, anonymised. Risk, evidence, recommendation, exactly as written.
See a full sample report - comment FINDING and I'll DM it.

09/10/2026

Your LL144 bias audit expires every year. Most of the ones published when the law landed in 2023 have quietly lapsed - and a lapsed audit is, for compliance purposes, no audit.

Two minutes on how to check yours: find the published summary, read its date, and count backwards twelve months. That's the whole test.

Comment SNAPSHOT - three seats this month for a 30-minute read.

09/09/2026

An audit dated 2024 is not a 2026 audit. Local Law 144 gives no grace period and no rollover — one year old, or it doesn't count.
Six slides on who's covered, what the bias audit actually measures, what has to be published, and the renewal date most employers never diarised.
Comment AUDIT and I'll DM you the LL144 checklist.

09/07/2026

Three seats this month for a 30-minute read of your AI risk. No pitch, no fee, and you keep the notes whether or not we ever work together.
We name the AI systems you actually run, map each one to the frameworks that apply, and flag the top two or three risks with evidence. If nothing serious turns up, I'll tell you that too.
Comment SNAPSHOT and I'll send the booking link.

09/04/2026

Regulators aren't asking whether your AI is perfect. They're asking whether you tested it — and that's a much harder question to bluff.

Ninety seconds on the difference between a model that happens to be fair and a model you can prove you checked, using the fair-lending case that closed this year: a $2.5 million settlement announced by the Massachusetts Attorney General on 10 July 2025, where the allegation wasn't just what the model did — it was that nobody had tested it.

Read the source: https://www.mass.gov/news/ag-campbell-announces-25-million-settlement-with-student-loan-lender-for-unlawful-practices-through-ai-use-other-consumer-protection-violations

Comment AUDIT and we'll send you the teardown.

09/02/2026

The Massachusetts AG didn't allege that a lender's AI was biased. It alleged the company never checked — and that's the finding that should worry you.

A $2.5 million settlement with a student lender, announced 10 July 2025. The AG named two things the model did — a Cohort Default Rate variable and an automatic denial rule keyed to immigration status — and one thing the company didn't do: test for disparate impact.

Six slides, every claim from the AG's published statement.

Read the source: https://www.mass.gov/news/ag-campbell-announces-25-million-settlement-with-student-loan-lender-for-unlawful-practices-through-ai-use-other-consumer-protection-violations

Comment AUDIT and we'll send you the teardown.

Address

Greenwood, IN

Telephone

+13175055697

Website

https://idharma.us/audit

Alerts

Be the first to know and let us send you an email when iDharma posts news and promotions. Your email address will not be used for any other purpose, and you can unsubscribe at any time.