AI Architect · Sovereign AI for regulated banking · LLM red teaming before go-live · https://teapawt.org/t.co/RJq8xP5EKb
Joined April 2015
- Tweets208
- Following253
- Followers66
- Likes518
Fake news beats prompt injection!
In an LLM trading agent benchmark, fabricated news hit 62% attack success. Prompt injection hit 14%.
Banks must test fake evidence, NOT just injected instructions. The agent trusts the format, not the fact.
huggingface.co/blog/Ercin/pr…
Big moment here. Abu Dhabi and OpenAI will make a set of government rules machine readable to help officials check compliance. Then AI reads the translation, not the law. Test it against the legal text. Version it. Name an owner. Traceable is NOT faithful!
ercin.info/ai-architect-duba…
Mark my words: one turn is NOT a pass! In a 989 prompt financial services benchmark, four more turns broke a median 89% of the prompts that survived turn one, across nine models. Red team bank chatbots over several turns. Turn one is the warmup.
huggingface.co/blog/Ercin/ll…
Two guardrails. The same 75 clean HarmBench copyright prompts. Granite Guardian scores 0. Qwen3Guard-4B scores 99. "Neither is broken." They define unsafe differently! No grading definition, no bank go-live evidence. Ask for it.
huggingface.co/blog/Ercin/gu…
Let's be honest, AI disclosure is part of the build! The CBUAE AI note says banks should be transparent when customers are interacting with an AI application. In both Arabic AND English. Simple as that. UAE banks, does your bot say it's a bot in both languages, every time?
rulebook.centralbank.ae/en/r…
ercin.info/cbuae-ai-guidance… is the only place the note names chatbots: institutions should ensure "promotional materials and chatbots comply with disclosure requirements." So test the disclosure line like any other output.
Pay attention to this number: 52% of DFSA Authorised Firms in the DIFC now use AI. And 21% still lack clear accountability or oversight mechanisms. That's not a model problem. It's an ARCHITECTURE problem! DIFC leaders, who can stop your system today?
661 firms, 88% participation, run June 2025. That's the DFSA AI Survey 2025. An AI Architect decides who can stop the system, what it may read and do, and how you prove it is safe. Accountability goes in before launch.
dfsa.ae/news/new-dfsa-ai-sur…
ercin.info/ai-architect-duba…
Folks, ask your AI consultant ONE question: who holds the switch-off?
If the answer is "just call us," at 2 a.m. on a public holiday? WRONG! The switch is THEIRS.
Look, this question maps to section 6(f) of the CBUAE AI guidance note.
Bank CIOs and CROs: who holds YOURS?
BTW, that's question h of ten. My buyer's guide for any AI consultant UAE banks meet: ten questions, five red flags, each question mapped to a clause. Nothing for sale. Frankly, EVIDENCE only. HUGE!
ercin.info/ai-consultant-uae…
CBUAE note, issued 11 Feb 2026:
rulebook.centralbank.ae/en/r…
Here's the thing: the ECB gave significant banks until 31 October 2026 to send their supervisors an action plan against AI-enabled cyber threats. Annex 1, area 2: "Accelerate vulnerability and patch management at scale." Patch speed is now a board question. NOT a ticket queue.
Source: ECB Banking Supervision letter SSM-2026-0301, 7 July 2026. Annex 1 runs from attack surfaces to crisis exercises: bankingsupervision.europa.eu…
Look. Your payments agent saw "KYC passed". It did NOT run KYC. It can't verify it. The money moved anyway. SAD!
That's Borrowed Clearance.
Signed, payload-bound clearance is NEVER inherited. Can't prove the check? Then you don't have it. STOP!
Folks, what it is, how to red-team for it, and the fix:
ercin.info/borrowed-clearanc…
LOOK at footnote 3 of SR 26-2. That's the US model-risk guidance (17 Apr 2026) that REPLACED SR 11-7.
Generative and agentic AI? "Not within the scope of this guidance." OUT!
So who tests your bank's LLM? BTW, the same footnote points back to your own risk management. YOU do.
SR 26-2: federalreserve.gov/supervisi…
Footnote 3 is in the attached guidance: federalreserve.gov/supervisi…
What testing an LLM before go-live looks like: ercin.info/llm-red-teaming-b…
US banks are BLOCKING third-party AI agents from customer accounts. HUGE.
Why? Some customers hand the agent their OWN login. The bank can't tell human from bot.
Folks, that's not an AI problem. It's IDENTITY. Every agent needs its own scoped credential. Sharing yours? WRONG!
Look, the fix is boring and it works: agents get their OWN identity, scoped permissions and an audit trail. Reuters via Times of India, 23 Oct 2026: timesofindia.indiatimes.com/…
"In-country inference" is NOT "everything stays in the UAE." Read the fine print!
OpenAI's own page: GPU inference stays in region. Auth, routing, analytics and some CPU work may run outside it.
Look, your bank's data-flow diagram needs BOTH boxes. Every time!
OpenAI, data residency and inference residency (see "Processing that may occur outside the selected region"): help.openai.com/en/articles/…
Context, Emirates NBD x OpenAI, 19 Oct 2026: emiratesnbd.com/en/media-cen…