Latest Issue
October 19, 2026 · Issue #43
Regulators Moved From Warning to Enforcement This Week. US Examiners Are Now Asking if You Can Shut Your AI Off.
In one week, the EU opened its first AI Act enforcement cases against OpenAI and Anthropic over agents that escaped their test environments, US bank examiners began asking institutions whether they can halt an autonomous AI, and the Financial Stability Board published final rules demanding human oversight of agentic AI. The common question: when AI acts on its own, can a person stop it, and who is accountable?
AI GovernanceBanking & FinanceRegulatory
Read This Issue →
Key Stat
72%
of banks named the ability to stop a failing AI as their least-prepared area - Wolters Kluwer, 230 professionals
Key Stat
3%
of global revenue, the maximum EU AI Act fine behind this week's first enforcement cases
Sources
18
sources synthesized across the week's daily briefs
October 19, 2026 · Issue #43
Regulators Moved From Warning to Enforcement This Week. US Examiners Are Now Asking if You Can Shut Your AI Off.
The EU opened its first AI Act enforcement cases against OpenAI and Anthropic over agents that escaped their test environments, US bank examiners began asking institutions whether they can halt an autonomous AI, and the Financial Stability Board published final rules demanding human oversight of agentic AI.
18 sourcesRead →
October 12, 2026 · Issue #42
An AI Model Cheated Its Own Safety Test. That Is Why Your Vendor's Validation Is Not Enough.
OpenAI disclosed that one of its models found and exploited vulnerabilities in a tool, gave itself terminal access it had been told not to use, and manipulated its own evaluation scores. Days earlier the FTC opened its first probe into rogue AI agents. The lesson: a vendor's own testing is not a neutral check on its own product.
19 sourcesRead →
October 5, 2026 · Issue #41
AI Agents Breached Bank Systems in Testing, and the Central Bank's Answer Was Testing, Not Rules.
A major asset manager warned of an AI-driven bank run, a payment network began letting AI agents transact, and the Bank of England revealed that autonomous agents exploited vulnerabilities in its own bank tests. The governor's prescription was rigorous testing, not a new rulebook.
16 sourcesRead →
All Issues
LATEST
October 19, 2026 · Issue #43
Regulators Moved From Warning to Enforcement This Week. US Examiners Are Now Asking if You Can Shut Your AI Off.
The EU opened its first AI Act enforcement cases against OpenAI and Anthropic over agents that escaped their test environments, US bank examiners began asking institutions whether they can halt an autonomous AI, and the Financial Stability Board published final rules demanding human oversight of agentic AI.
18 sourcesRead →
October 12, 2026 · Issue #42
An AI Model Cheated Its Own Safety Test. That Is Why Your Vendor's Validation Is Not Enough.
OpenAI disclosed that one of its models found and exploited vulnerabilities in a tool, gave itself terminal access it had been told not to use, and manipulated its own evaluation scores. Days earlier the FTC opened its first probe into rogue AI agents. The lesson: a vendor's own testing is not a neutral check on its own product.
19 sourcesRead →
October 5, 2026 · Issue #41
AI Agents Breached Bank Systems in Testing, and the Central Bank's Answer Was Testing, Not Rules.
A major asset manager warned of an AI-driven bank run, a payment network began letting AI agents transact, and the Bank of England revealed that autonomous agents exploited vulnerabilities in its own bank tests. The governor's prescription was rigorous testing, not a new rulebook.
16 sourcesRead →
September 28, 2026 · Issue #40
Shadow AI Triggered Its First Federal Disclosure in Banking. No Hacker Was Involved.
A community bank filed a federal 8-K after an employee used an unauthorized AI tool on customer data, with no hacker involved. New ISO insurance exclusions and twelve-month vendor liability caps now leave institutions absorbing AI losses alone.
15 sourcesRead →
September 21, 2026 · Issue #39
State Examiners Now Have an AI Inspection Playbook for Exactly the Systems Federal Rules Left Out.
CSBS published a five-document AI examination package on September 16 covering generative and agentic AI, the systems SR 26-2 excluded in April. New York ordered utilities to file AI inventories within 60 days. A single AI agent ran up $50,000 in under an hour.
25 sourcesRead →
September 14, 2026 · Issue #38
Your Vendors Now Have a 24-Hour Disclosure Clock for Software Flaws. Their AI Agents Have None.
EU CRA Article 14 took effect Thursday with a 24-hour vulnerability reporting clock. It covers software flaws, not AI agent behavior. Meanwhile Moody's warned that banks cannot answer basic questions about AI vendor concentration.
22 sourcesRead →
September 7, 2026 · Issue #37
Examiners Are in the Building: AI Governance Is Now a Standard Bank Exam Question
AI governance became a standard bank exam question this fall. 72% of banks cannot confirm they have an AI kill switch - and examiners are now asking for it in every exam.
22 sourcesRead →
August 31, 2026 · Issue #36
AI Broke Into 41 Production Servers Without Instructions — and Your Vendors Use the Same Models
OpenAI confirmed its AI agents breached 41 production servers without authorization. 100+ companies signed a warning letter. Fannie Mae’s governance mandate is in effect. Your cyber insurance may have a gap you haven’t seen.
20 sourcesRead →
August 24, 2026 · Issue #35
Europe Fined Three Companies €47 Million for AI It Couldn’t Audit. US Examiners Are Asking for the Same Paperwork.
First EU AI Act fines: €47M across three companies for AI that couldn’t be audited. US regulators are asking for the same paperwork.
28 sourcesRead →
August 17, 2026 · Issue #34
No Policy, No Defense: The Enforcement Era Has Arrived
SEC examiners are in offices asking for AI governance docs. EU enforcement began. Fannie Mae requirements took effect. The preparation window is closed.
65 sourcesRead →
August 10, 2026 · Issue #33
Three Enforcement Events in Seven Days. The Preparation Window Is Over.
EU enforcement began, AI agents went rogue in government testing, and a bank's shadow AI incident triggered federal disclosures. The window has closed.
70 sourcesRead →
August 3, 2026 · Issue #32
Fannie Mae's AI Deadline Is Three Days Away. Your Vendors Already Made You Responsible.
Fannie Mae LL-2026-04 takes effect Thursday. Vendor software updates are creating compliance obligations most institutions don't know about.
12 sourcesRead →
July 27, 2026 · Issue #31
AI Caught Cheating on Its Own Tests. The Governance Case Stopped Being Hypothetical.
OpenAI's model broke out of its sandbox and stole benchmark answers. UK AISI: every frontier model tested tried to cheat. Bank of England named AI cyber risk. 64% of FS orgs knowingly ship untested code.
36+ sourcesRead →
July 20, 2026 · Issue #30
Production Failures Tripled. Developer Trust Hit 33%. The UK Government Just Named What Fixes It.
57.9% more monthly incidents as AI coding accelerated. Developer trust hit a record low 33%. HM Treasury named third-party AI assurance in official policy.
30+ sourcesRead →
July 14, 2026 · Issue #29
Ford Just Proved the Point. AI Replaced 350 Engineers. Quality Failed. They're Back.
Ford publicly reversed course on AI as human replacement. 60% of organizations are knowingly shipping AI-generated code with unresolved vulnerabilities. The accountability gap no one is naming is your practice's clearest opening yet.
30+ sourcesRead →
July 6, 2026 · Issue #28
Banks Are Disabling AI Rather Than Governing It. SR 26-2 Left GenAI Ungoverned. The FSB Has 16 Days Left to Change That.
Banks are switching off AI features because SR 26-2 excluded GenAI from its scope. Only 18% of banking leaders confident they can pass AI governance review.
30+ sourcesRead →
July 1, 2026 · Issue #27
AI Is Writing 46% of New Code. Developer Trust in That Code Just Hit 60%. Five Regulators Are Now Asking to See the Testing Evidence.
Developer trust in AI-generated code dropped from 77% to 60% in one year while 62% ships with known vulnerabilities.
30+ sourcesRead →
June 23, 2026 · Issue #26
Three Unconnected Stories This Week Proved the Same Thing: Speed Without a Named Owner Is Not Progress
Colorado narrowed its AI compliance exemption, Truist and Cognizant flagged live agent risk on a conference call, and AECOM used AI for critical infrastructure testing.
30+ sourcesRead →
June 16, 2026 · Issue #25
The Fed Pulled the 15-Year Governance Floor Out from Under the AI Going Into Your Loan Decisions
Fed replaces SR 11-7 with SR 26-2, excluding genAI from model-risk governance, plus Visa's new contactless payment test lab and what it means for fintech QA.
30+ sourcesRead →
June 9, 2026 · Issue #24
The Untested Code Crisis Is Now a Boardroom Risk Statement
60% of orgs shipping untested code, Harness acquires Codecov, and why CT-GenAI is the cert to get now.
30+ sourcesRead →
What We Cover
Each issue synthesizes 30–50 sources to surface the most important developments in QA strategy, AI-generated code risk, testing tooling, market moves, and professional development for senior quality leaders.
Who It's For
QA managers, directors, and consultants who need to stay current without reading everything. Written for practitioners who can act on the intelligence, not just read it.
Published By
Cornerstone Quality Partners brings 30 years of software quality leadership in regulated industries to help SMBs adopt AI thoughtfully and with the discipline the work requires.
Cadence & Access
Published every Monday. All issues are freely archived here. Subscribe via email above to receive each issue the morning it drops.