In partnership with

THE AI GRC DESK

Practical AI for GRC practitioners. Using it, governing it, and building it responsibly.

The AI GRC Desk: AI Governance and Security Round-up. Regulation, supplier incidents and key deadlines. Week ending 15 November 2026.

AI Governance and Security Round-up · Week ending 15 November 2026

Hello Hello,

Keeping up with AI governance means following several conversations at once.

There are regulatory publications to interpret, consultations to respond to, supplier incidents to assess and implementation deadlines approaching. Each needs a different response.

This week brings all four together.

The European Commission’s Article 50 guidance reached the Official Journal. New disclosures added detail to the Fireworks AI incident and its downstream impact. In the UK, an education-sector consultation opened while the ICO’s agentic AI call for evidence approaches its closing date.

Here is what matters, what the evidence supports and what I would take into the next governance conversation.

THE ONE THING TO TAKE AWAY

Give every development a clear status before deciding what to do about it.

Is it a binding requirement, guidance on an existing duty, a consultation, a reported recommendation or an incident disclosure?

Then establish who it applies to, what changed and who owns the response.

That small discipline helps turn a long watchlist into useful decisions.

01 · EU AI Act: Article 50 guidance reaches the Official Journal

Status: Published Commission guidance on existing legal obligations.

The Commission’s guidelines on Article 50 transparency obligations were published in the Official Journal on 12 November, under reference C/2026/6005. The Commission’s library page first published the guidance in July and was updated on 5 November. This is an important distinction when explaining what is new this week.

Article 50 covers transparency obligations for certain AI systems, including interactions with people and marking or disclosure requirements for AI-generated content. The Commission states that these obligations have applied since 2 August 2026. The guidelines support interpretation and implementation; they do not create a separate set of legislative duties.

Why it matters

A policy saying “we disclose AI use” leaves several practical questions unanswered.

Which systems require disclosure? Who provides the notice? Which content needs marking? What evidence shows the requirement has been implemented?

Action for security and governance teams

  • Reference the adopted guidance in your applicability assessment.

  • Separate provider responsibilities from deployer responsibilities.

  • Review customer-facing AI interactions and synthetic-content workflows.

  • Record the owner and evidence for each applicable obligation.

The Commission’s current implementation timeline identifies 2 December 2026 as the Article 50(2) transitional deadline for certain providers of synthetic-content-generating systems already placed on the market before 2 August. It also lists new prohibitions relating to non-consensual sexual deepfakes and child sexual abuse material for that date. Scope matters: this is not a blanket extension for every Article 50 obligation.

02 · Fireworks AI: the supplier incident with downstream consequences

Status: Vendor-reported incident findings; investigations remain ongoing.

The underlying incident began in October. This week’s significance is the additional disclosure about affected data and downstream access.

Fireworks’ 9 November update confirms impacts involving stored secrets and, for certain customers, datasets and incomplete model weights. It also identifies account and billing information among the affected categories. Fireworks says it has not detected impact to its Inference, Training SDK and Nexus services, which it describes as zero data retention by default. Those statements remain qualified by its ongoing investigation.

Notion’s disclosure

Notion says fewer than 1% of workspaces were affected. Its advisory identifies Notion AI chat transcripts used for the Insights feature in the affected dataset.

Notion disconnected from Fireworks, disabled Insights during the investigation and contacted affected customers. It also revoked exposed Notion API keys and reported exposed third-party keys to their issuing vendors. Its advice to affected customers includes reviewing the available data and rotating secrets that appeared in affected conversations.

Cursor’s disclosure

Cursor’s 13 November update describes an attacker obtaining a key issued to Fireworks and using it to copy historical coding-agent traces from a read-only storage bucket.

Some traces contained unredacted credentials. Cursor says those credentials enabled further access to selected internal services. A stolen credential was subsequently used to place malicious Claude Code-branded extensions on its marketplace cache. Cursor states that Anthropic did not publish those copies and that this was not a vulnerability in Anthropic’s product.

Why it matters

For me, this is a reminder that an AI supplier review needs to follow the data beyond the product interface.

Prompts, traces, support messages and evaluation datasets can create additional copies of sensitive information. Credentials inside those copies can create further access paths.

Action for security teams

  • Map inference, evaluation, analytics and support data flows.

  • Establish which subprocessors receive each category of data.

  • Verify retention arrangements for the specific feature or endpoint.

  • Check how secrets are detected and removed from stored conversations and traces.

  • Test how quickly vendor access can be revoked and affected credentials identified.

  • Include extension distribution and supplier-held datasets in incident planning.

A useful supplier question is:

“If your provider is compromised, how will you establish whether our data or credentials were involved?”

03 · UK: accountability for customer-facing agents remains with the business

Status: Existing-law guidance, updated outside this week’s window.

The CMA updated its guidance on consumer law and AI agents on 30 October.

Its message is direct: businesses remain responsible for what their AI agents do, including when another organisation designed or supplies the agent. The guidance covers transparency, testing, monitoring and correcting problems. It also warns that consumer-law breaches can attract fines of up to 10% of worldwide turnover.

The DRCF’s 30 October joint statement brings together the CMA, FCA, ICO and Ofcom’s work on personal AI agents and the application of existing regulatory frameworks.

Why it matters

A supplier contract is only one part of accountability.

If an agent handles refunds, recommends products or communicates terms, someone inside the business still needs to own its behaviour and the customer outcome.

Action for security and governance teams

Choose one customer-facing agent and check:

  • What decisions can it make?

  • Which customer rights must it preserve?

  • When does a person intervene?

  • How are incorrect decisions detected and corrected?

  • Who can stop the service?

An immediate opportunity to contribute

The ICO’s agentic AI call for evidence closes at the end of 20 November 2026. It covers security, transparency, accountability, automated decision-making, fairness and lawful processing.

If your team has practical evidence about permissions, oversight or unexpected data use, this is an opportunity to help shape regulatory guidance. Use approved, anonymised examples.

Your Next Side Hustle is Already on This List

Ready to stop scrolling job boards and start building something of your own?

The Hustle's free "Side Hustle Ideas Database" uncovers 100 unconventional, actually-actionable ways to earn on the side.

Inside you'll find:

  • A curated list of 100 real side hustle ideas spanning e-commerce, services, content, and weird corners of the internet most people never think to look

  • Opportunities at every level, whether you have 5 hours a week or 50, zero capital or a small runway

  • Offbeat, under-the-radar ideas you won't find in generic "how to make money" listicles

Subscribe for free today and get instant access. Your next income stream is one click away.

04 · England: proposed safety certification for education AI

Status: Consultation, not an implemented certification requirement.

On 10 November, the Department for Education opened a consultation on a proposed product safety certification scheme for EdTech and AI used in schools and other education settings in England.

The consultation closes on 4 February 2027. It seeks views on providing meaningful safety assurance while keeping the approach workable for suppliers and education settings.

Why it matters

For suppliers entering education, evidence of safety may increasingly influence procurement and market access.

My interpretation: this is another reason to build assurance into product development early. Retrofitting evidence when a customer asks for it is harder.

Action for relevant teams

Review the proposal with product, safeguarding, privacy and security owners. Identify what evidence you could already provide and where testing or ownership remains unclear.

05 · US federal policy: recommendations require careful labelling

Status: Reported policy recommendation; no new safe harbour verified here.

Cybersecurity Dive reports that the White House’s Super Intelligence Force discussed cybersecurity at its 12 November meeting.

According to that reporting, the group recommended limitations on regulatory liability for good-faith voluntary reporting of vulnerabilities and incidents, where lawful and practical, for at least 12 months. I could not independently retrieve the official readout, so this account relies on secondary reporting.

What this means in practice

This reporting does not establish an operative exemption your organisation can rely on.

Before changing incident-response assumptions, identify the relevant agency, the published instrument and its conditions.

Separately, AP reporting carried by Euronews describes an informal reporting-line change involving CAISSI, the US government AI testing body. The account relies on former officials and says the White House did not answer questions about the organisation. Treat this as an institutional development to monitor, with official confirmation outstanding.

06 · Standards watch: continue the work while revisions develop

Status: Current published resources; revision activity to monitor.

ISO continues to list ISO/IEC 42001:2023 as its published AI management-system standard. NIST’s AI RMF page states that AI RMF 1.0 is being revised as part of the White House AI Action Plan. The pages checked did not establish a replacement edition released this week.

My takeaway

Keep improving ownership, inventories, risk assessments and control evidence. Record which framework version your programme uses and assess changes when authoritative material becomes available.

07 · Threat context: the NCSC’s assessment to 2028

Status: Earlier assessment, retained for context.

The NCSC’s assessment of AI and cyber intrusion considers how the threat could develop to 2028. Its judgements include increased state use of AI-enabled cyber capability and continued exposure through known weaknesses in critical infrastructure.

It also assesses AI-powered malware being used in a ransomware attack on UK critical national infrastructure by 2028 as highly likely. These are intelligence assessments and scenarios, rather than confirmation that every described outcome has occurred.

Action for security teams

Use the assessment to challenge existing plans: vulnerability remediation, legacy-system exposure, supplier access, recovery and incident-response exercises.

For organisations operating OT or critical infrastructure, involve operational owners in that discussion.

DATES TO KEEP IN VIEW

These dates have different scopes. Confirm applicability before adding an obligation to your register.

Date

Development

Who should assess it

20 November 2026

ICO agentic AI call for evidence closes.

Teams contributing practical evidence on agentic AI and data protection.

2 December 2026

Article 50(2) transition for certain existing systems; additional EU prohibitions.

Relevant AI providers and legal owners.

1 January 2027

Colorado ADMT and Chatbot Safety Acts take effect, according to the Attorney General’s current page.

In-scope developers, deployers and chatbot operators.

1 January 2027

California ADMT compliance requirements begin for covered significant decisions.

Businesses using in-scope automated decision-making.

1 January 2027

New York registration requirement begins for large frontier developers.

Organisations meeting the statutory developer definitions.

4 February 2027

England’s EdTech and AI safety certification consultation closes.

Education suppliers, procurers and assurance teams.

Also on the watchlist: New York’s youth chatbot legislation, signed on 28 October. The Governor’s announcement confirms restrictions on certain features offered to minors, but describes implementation as occurring “in the coming months”. I have not verified the precise commencement date from the enacted text, so I am leaving it outside the confirmed deadline table.

For longer-term EU planning, the Commission’s current timeline lists 2 December 2027 for Annex III high-risk rules and 2 August 2028 for high-risk AI embedded in Annex I regulated products, reflecting Digital Omnibus amendments.

QUESTIONS TO TAKE INTO YOUR NEXT MEETING

  • Which development changes an obligation, control or decision for us?

  • Can we trace where our AI suppliers send and retain our data?

  • Who owns upcoming requirements, and what evidence demonstrates readiness?

  • Could we identify and revoke supplier-held credentials quickly?

  • Which watchlist items still need authoritative confirmation?

My practical suggestion this week: choose one relevant update and record its status, applicability, owner, action and evidence.

What would need to change in your organisation because of this update, and who is responsible for making it happen?

Princess David Okoro, CISM
Author, The AI GRC Desk
Find me on LinkedIn: Princess David Okoro
Subscribe to The AI GRC Desk: aigrcdesk.com/subscribe

The practitioner's guide to AI in GRC. Using it, governing it, and building it responsibly.