AI Agents Start Making Your Calls, and You're Liable (Nov 5, 2026)

🎧 Listen ad-free:

🌐 Daily Briefing: https://djamgamind.com/daily
📚 Research Hub: https://djamgamind.com/pdfs

🔍 Keywords: AI Agents, AI Liability, AI Governance, Grok Bot, Air Canada Chatbot, Anthropic FlashProve, Provable Inference, Data Centers, Supply Chain Security, Cybersecurity

Today’s question: AI now acts in your name. Can you prove what it did, and who answers for it?

Summary:
Grok Bot now places real phone calls for paying users, and its terms make the user the “caller of record.” Anthropic released FlashProve, which proves cryptographically that an answer came from the model a provider claims. Voters backed limits on data centers in 38 of 39 completed local measures, and Goldman Sachs says the states likely to restrict them hold about 45% of planned U.S. capacity. And the hackers behind Europe’s freezer attacks reached French blood banks through a supplier’s supplier.

Important Topics:

• The agent picks up the phone: Grok Bot calls, and who’s liable.

• Proving what AI did: FlashProve, AI-media detection, AI-designed hardware.

• The trust deficit: the data-center vote and Goldman’s 45%.

• The back door: blood banks hit through a third-party component.

• Today’s corners: Cert Corner, Red Flag or Green Light?, and the Boardroom Call.

📞 The agent picks up the phone

The Rundown AI · TeslaNorth · TBPN

• Grok Bot calls: SpaceXSI’s agent can now sit on hold, book appointments and negotiate bills on U.S. numbers. If a rep asks something it can’t answer, you get a message mid-call and type the reply (The Rundown AI).

• You’re on the hook: the terms make users the “caller of record” and “solely responsible” for calls and phone laws; the bot discloses it’s a “virtual assistant” (The Rundown AI).

• Scale: Grok Voice already answers 20,000 Starlink calls a day, by SpaceXSI’s count (The Rundown AI).

• Agents managing agents: Meta acquired Conductor, which helps developers run many AI agents at once (Axios AI+ exclusive). OpenAI is buying Pierre to adapt Git “to how teams work with coding agents” (The Rundown AI; TBPN).

Why it matters: When an agent speaks for you, its words are yours. In Moffatt v. Air Canada (2024), a British Columbia tribunal held the airline liable for wrong refund information its chatbot gave a customer. Decide what your agents may promise before they talk to anyone, and log every conversation.

🤝 PARTNER SLOT
This space is reserved for one enterprise AI partner each day: agent governance, AI security and accountability for the leaders who read this briefing. Sponsorship options and media kit →

🔏 Proving what AI did

The Deep View · TBPN

• FlashProve: Anthropic’s open-source system uses cryptography to prove an answer came from the model a provider claims, not one quietly swapped in by a router. Anthropic: a user today “cannot check that the responses came from the specific model that a developer claims is responding” (The Deep View).

• The proof checks “in seconds on a CPU” and doesn’t reveal the model’s weights (The Deep View).

• Spotting AI media: Pangram launched image and video detection, claiming 99.96% accuracy on images and 98.57% on video (company figures via TBPN).

• AI-designed hardware: OpenAI says GPT-6 Astra designed a test device end to end, but it still missed details like the manufacturer’s production limits (The Deep View).

The Number: 212 million. Tokens per employee per day on OpenAI’s hardware teams, up 23-fold since February (The Deep View).

Why it matters: “Which model did this?” is becoming an audit question. Ask your vendors how they prove which model served each request.

🗳️ The trust deficit

The Deep View · The Rundown AI · Axios AI+ · The Next Web · TechCrunch

• The tally: “restrict” results won 38 of 39 completed data-center measures (Redbud via The Deep View). In NBC’s exit poll, 76% would oppose a data center near them and 57% expect AI to do more harm than good (The Rundown AI).

• Goldman Sachs: states likely to restrict data centers “account for roughly 45% of planned US data center capacity”; Texas alone is about 25% (The Next Web).

• Next in Congress: a Sanders and Ocasio-Cortez bill would pause new data centers nationwide; it could face a veto (TechCrunch). Sen. Ed Markey: “They are going to need to plant new forests for all of the subpoenas we’ll be printing.”

• How labs adapt: smaller 20 to 30 MW sites and more building overseas, such as Google’s $15 billion in Malaysia (The Deep View).

• The new deal voters want: companies pay for their own infrastructure and power, tax breaks face repeal, and no more nondisclosure agreements. OpenAI’s Chris Lehane: “A lot happens in two years.” (Axios AI+)

Why it matters: Capacity will move to smaller sites and new regions. Your contracts should say where your workloads run.

🩸 The back door

CBS News

• Hackers raised temperature limits at three French plasma cold-storage sites through refrigeration supplier Coolworld, whose flaw “relates to a component from a third-party supplier” (CBS News).

• Independent monitoring caught it fast: “no blood products or samples were compromised” (EFS).

• The October supermarket attacks may cut Ahold Delhaize’s fourth-quarter sales by up to €250 million (CBS News).

Why it matters: You’re only as secure as your supplier’s supplier. Keep alarms that don’t depend on the vendor’s own system.

🎯 Today’s corners

🎓 Cert Corner: A cold-storage vendor’s monitoring system is hacked through a component from its own supplier. What best limits the damage at your sites?

A) Put the equipment on its own network segment with independent alarms

B) A stronger non-disclosure agreement

C) Wait for the vendor’s patch

D) Turn off local alarms

🚩 Red Flag or Green Light? Your collections team wants an AI agent to call customers about overdue invoices. It says it’s a virtual assistant, every call is transcribed, a human can take over anytime, and it can’t offer discounts on its own. 🚩 or ✅?

🏛️ The Boardroom Call: Your AI agent promised a supplier a 15% discount it wasn’t allowed to give, and the supplier accepted.

A) Honor it

B) Void it

C) Honor it once, then cap what agents can promise

D) Make the agent’s vendor pay

Answers below. Tell us yours in the comments.

Strategic C-Suite Takeaways

1. Decide what your agents may promise before they talk to anyone.

2. Log every agent call and conversation.

3. Ask vendors how they prove which model answered.

4. Know where your AI workloads run.

5. Keep monitoring independent of your suppliers.

What Else Happened in AI on November 5, 2026?

• Canva cut 1,271 jobs, about 22% of staff. TBPN

• Apple is cutting its engineers’ internal AI use to save money. The Deep View

• Oracle scrapped a $7 billion chip deal with Tencent under U.S. pressure. The Deep View

• Google will retire Gemini 2.5 Pro and Flash. The Deep View

• Ricursive released Ripple-65B Preview, built for $32 million. The Rundown AI

• Anthropic product executive Cat Wu is moving to OpenAI to lead product management for its core services. Axios AI+

✅ Our take on today’s corners

🎓 Cert Corner: A. Segment the equipment and keep alarms that don’t rely on the vendor’s system; that’s what caught the blood-bank attack in time. Practice thousands more in DjamgaTech.

🚩 Green Light, with conditions. Disclosure, transcripts, a human who can step in and no power to discount are the right controls. Check local phone and collection rules and offer an easy opt-out. 📘 AI security and governance controls are a full domain of the AWS AI Practitioner exam: our 2026 guide.

🏛️ C. Honor it once, then cap what any agent can commit to without a human. After the Air Canada ruling, voiding it is risky and blaming the vendor rarely works. Your org would do it differently? Tell us. 📘 Setting up AI governance is a core topic of Google’s Gen AI Leader exam: our 2026 guide.

🚩 Red Flag or Green Light is open for one sponsor. AI security companies: djamgamind.com/partners

🤝 Advertise with AI Unraveled & DjamgaMind

Put your enterprise AI brand in front of the CTOs, CISOs and engineering leaders who make AI buying decisions, daily, in English and French, on podcast apps, YouTube and Substack. Start by sponsoring one daily corner for four weeks.
Sponsorship options & media kit → · Email: etienne_noumen@djamgamind.com

🔗 Tools from the show

• DjamgaTech: Pass your AWS, Azure, GCP & AI certification exams.

• 📘 New eBook: Google Cloud Generative AI Leader Exam Guide (2026), the executive guide, updated for the new Gemini product names, with 60 practice questions: https://djamgamind.com/ebooks/?book=gcp-gail&ep=20261105&src=substack

• 💼 Google Workspace for your team (business email, Docs, Drive, Gemini). Free 14-day trial: https://referworkspace.app.goo.gl/Q371 or https://sharegoogleworkspace.advocate.impact.com/mQZcXGY. 10% off your first year (Business Starter or Standard, Americas) with code 63F733CLLY7R7MM, 63F7D7CPD9XXUVT, 63FLKQHWV3AEEE6, 63JGLWWK36CP7WM. More codes: https://djamgamind.com/toolkit/#google-workspace

• All our apps and tools: AI & ML Tutor PRO, iQz, DjamgaMind Kids and more.

🎧 Prefer it commercial-free? Subscribe on Apple Podcasts.

⚗️ PRODUCTION NOTE: We Practice What We Preach.
AI Unraveled is researched and produced by Etienne Noumen, P.Eng. (Calgary, Canada) using a hybrid “Human-in-the-Loop” workflow.