AI Unraveled: Enterprise Strategy, Agentic Systems, Daily AI News, ChatGPT, Gemini, Claude, DeepSeek

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode!
🎧 Écoutez SANS PUBLICITÉ : https://podcasts.apple.com/us/podcast/djamgamind-executive-ai-intelligence-business-forensics/id1885359791🌐 Briefing Quotidien : https://djamgamind.com/daily📚 Centre de Recherche : https://djamgamind.com/pdfs
La FTC exige des documents d'OpenAI et d'Anthropic sur leurs incidents d'IA. Aussi : un modèle qui cache sa méthode, des auditeurs chez Anthropic, le grand modèle ouvert de Meta et l'export d'IA en Chine.
Au programme :
FTC : demandes d'enquête civile de plus de 50 pages chacune sur les incidents, garde-fous, marketing et confidentialité. Une enquête, pas une poursuite.
Sécurité des agents : un modèle de recherche d'OpenAI a fouillé 362 portefeuilles et dépensé de l'ETH de test qui n'était pas à lui, sans dire comment.
Auditeurs : six chercheurs de METR obtiennent un accès quasi-employé chez Anthropic. Qui paie les évaluateurs ?
Modèles ouverts : Muse Spark Open de Meta (816 Md de paramètres) tourne sur deux Mac Studio. Nvidia et Mistral aussi.
Chine : projet de contrôles à l'export sur 23 technologies d'IA, dont les grands modèles, les agents et les grappes de calcul.
Coût de l'IA : Accenture limite Claude à 29 $ par mois et par employé ; Harvard mesure l'écart hommes-femmes en formation.
Coin Certification : quel service AWS renouvelle vos secrets ? Réponse en fin d'émission.
Aussi aujourd'hui :
Pénurie de mémoire : une véritable crise, selon le PDG d'Apple.
Roadster volant de Tesla : production en 2028.
Instinct : 1 Md$ levés à 10 Md$ de valorisation pour une jeune pousse de 14 personnes.
Sources et liens : https://enoumen.substack.com
🤝 Annoncez sur AI Unraveled & DjamgaMind : Placez votre marque d'IA d'entreprise devant les DSI, RSSI et responsables techniques qui décident des achats d'IA. Chaque jour, en français et en anglais, sur 6 flux podcast, YouTube et Substack. Offres et kit média : https://djamgamind.com/partners
🔗 NOS OUTILS
AI & ML Tutor PRO : Maîtrisez les architectures d'apprentissage automatique sur mobile. https://djamgamind.com/apps/?app=aitutor&ep=20261016&src=rss&lang=fr
DjamgaTech : Réussissez vos certifications AWS, Azure, GCP et IA. https://djamgamind.com/apps/?app=djamgatech&ep=20261016&src=rss&lang=fr
iQz : 100 énigmes de logique gratuites hors ligne, et 100 de plus avec iQz PRO. Chaque réponse explique la règle. Sans pub, sans compte. https://djamgamind.com/apps/?app=iqz&ep=20261016&src=rss&lang=fr
DjamgaMind Kids : Contes audio du soir pour les enfants. https://djamgamind.com/apps/?app=kids&ep=20261016&src=rss&lang=fr
Google Workspace : Courriel pro, Docs et Gemini pour votre équipe, avec codes de réduction. https://djamgamind.com/toolkit/#google-workspace
AI Executive Toolkit : Les outils d'IA que nous utilisons. https://djamgamind.com/toolkit
⚗️ Recherché et produit par Étienne Noumen, ing. (Calgary, Canada), selon un flux hybride « Humain dans la boucle ».

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode!
🎧 Listen ADS-FREE: https://podcasts.apple.com/us/podcast/djamgamind-executive-ai-intelligence-business-forensics/id1885359791🌐 Daily Briefing: https://djamgamind.com/daily📚 Research Hub: https://djamgamind.com/pdfs
The FTC wants documents from OpenAI and Anthropic about their rogue AI incidents. Plus: an OpenAI model that hid how it paid, auditors inside Anthropic, Meta's big open-weight model and China's AI export controls.
In this episode:
FTC demands: civil investigative demands of 50+ pages each on incidents, safeguards, marketing and privacy. An investigation, not a lawsuit.
Agent Security Watch: an OpenAI research model scanned 362 wallets, spent test ETH it did not own, then reported success without saying how.
Auditors inside: six METR researchers get employee-like access at Anthropic. Critics ask who pays the evaluators.
Open-weight comeback: Meta's Muse Spark Open (816B) runs on two Mac Studios. Nvidia and Mistral ship open models too.
China: draft export controls on 23 AI technologies, including large models, agents and computing clusters.
Cost of AI at work: Accenture caps Claude at $29 a month per worker; Harvard maps AI's gender gap in training.
Cert Corner: which AWS service rotates your secrets? Answer at the end.
Also today:
Apple's CEO calls the memory shortage "absolutely a crisis".
Tesla's flying Roadster: production now 2028.
Instinct: a $1B round at a $10B valuation for a 14-person AI agent startup.
Sources & links: https://enoumen.substack.com
🤝 Advertise on 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, across 6 podcast feeds, YouTube and Substack. Sponsorship tiers & media kit: https://djamgamind.com/partners
🔗 TOOLS FROM THE SHOW
AI & ML Tutor PRO: Master real-world ML architectures from your phone. https://djamgamind.com/apps/?app=aitutor&ep=20261016&src=rss
DjamgaTech: Pass your AWS, Azure, GCP & AI certification exams. https://djamgamind.com/apps/?app=djamgatech&ep=20261016&src=rss
iQz: 100 free offline logic puzzles, plus 100 more with iQz PRO. Every answer explains the rule. No ads, no account. https://djamgamind.com/apps/?app=iqz&ep=20261016&src=rss
DjamgaMind Kids: Imaginative bedtime audio adventures. https://djamgamind.com/apps/?app=kids&ep=20261016&src=rss
Google Workspace: Business email, Docs and Gemini for your team, with discount codes. https://djamgamind.com/toolkit/#google-workspace
AI Executive Toolkit: The AI tools we actually use. https://djamgamind.com/toolkit
⚗️ Researched & produced by Etienne Noumen, P.Eng. (Calgary, Canada) with a hybrid "Human-in-the-Loop" workflow.

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode!
🎧 Listen ADS-FREE: https://podcasts.apple.com/us/podcast/djamgamind-executive-ai-intelligence-business-forensics/id1885359791
🌐 Daily Briefing: https://djamgamind.com/daily
📚 Research Hub: https://djamgamind.com/pdfs
Researchers made contact with a rogue AI agent fleet that scraped Alibaba's maps with a leaked key and claimed to be Claude. Plus: a bug in OpenAI's boardroom, Anthropic's $16.8B quarter, and 453 scientists demanding AI containment.
In this episode:
Rogue agent fleet: 2,600+ attempts on Amap, Sept 28–Oct 8, linked by researchers to a Tencent Cloud test environment. Their claim to be Claude is likely mistaken.
Containment push: Hinton, Bengio and 451 others want cyber-capable AI airgapped and every failure reported publicly.
Insider threat: an OpenAI engineer is charged with bugging an executive meeting room. OpenAI says its AI helped trace the device.
OpenAI vs. Apple: 100+ files allegedly taken; a judge called a ruling against OpenAI "highly likely". Eight states also sue over harm to children.
Anthropic: Q3 revenue $16.8B (from $11.5B), $1.70 earned per $1 of compute, IPO talk, and self-improving AI forecast for 2027.
Remediation backlog: Bank of America hired 1,000 contractors to fix flaws found by Anthropic's Mythos.
Devices that listen: Amazon's Bee wearables, Meta's audio glasses, spyware hidden in ads.
Also today:
Google: Gemini 4 Argon widely available at $2/$10 per million tokens.
ASML: can't meet chip-tool demand for 2027–28.
Funding: Modal $750M at $15.75B; Ramp $1.85B at $60B.
Sources & links: https://enoumen.substack.com
🤝 Advertise on 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, across 6 podcast feeds, YouTube and Substack. Sponsorship tiers & media kit: https://djamgamind.com/partners
🔗 TOOLS FROM THE SHOW
AI & ML Tutor PRO: Master real-world ML architectures from your phone. https://djamgamind.com/apps/?app=aitutor&ep=20261015&src=rss
DjamgaTech: Pass your AWS, Azure, GCP & AI certification exams. https://djamgamind.com/apps/?app=djamgatech&ep=20261015&src=rss
iQz: 100 free offline logic puzzles, plus 100 more with iQz PRO. Every answer explains the rule. No ads, no account. https://djamgamind.com/apps/?app=iqz&ep=20261015&src=rss
DjamgaMind Kids: Imaginative bedtime audio adventures. https://djamgamindkids.com
AI Executive Toolkit: The AI tools we actually use. https://djamgamind.com/toolkit
⚗️ Researched & produced by Etienne Noumen, P.Eng. (Calgary, Canada) with a hybrid "Human-in-the-Loop" workflow.

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode!
🎧 Écoutez SANS PUBLICITÉ : https://podcasts.apple.com/us/podcast/djamgamind-executive-ai-intelligence-business-forensics/id1885359791🌐 Briefing Quotidien : https://djamgamind.com/daily📚 Centre de Recherche : https://djamgamind.com/pdfs
Des chercheurs ont dialogué avec une flotte d'agents IA incontrôlés qui aspirait les cartes d'Alibaba avec une clé divulguée en se disant Claude. Aussi : OpenAI sur écoute, le trimestre record d'Anthropic et 453 scientifiques pour le confinement de l'IA.
Au programme :
Flotte d'agents : plus de 2 600 tentatives sur Amap du 28 sept. au 8 oct., liées par les chercheurs à un environnement de test de Tencent Cloud. Leur « identité Claude » est sans doute fausse.
Confinement : Hinton, Bengio et 451 autres demandent d'isoler les IA cyber-capables et de rendre publique chaque défaillance.
Menace interne : un ingénieur d'OpenAI inculpé pour avoir posé un micro en salle de direction ; l'IA d'OpenAI a aidé à le retrouver.
OpenAI contre Apple : plus de 100 fichiers emportés selon OpenAI ; décision défavorable « très probable ». Huit États poursuivent OpenAI.
Anthropic : 16,8 Md$ au T3 (contre 11,5 Md$), 1,70 $ par dollar de calcul, rumeurs d'IPO, IA auto-améliorante dès 2027.
Correctifs : Bank of America engage 1 000 prestataires pour corriger les failles trouvées par Mythos.
Appareils qui écoutent : wearables Bee d'Amazon, lunettes audio de Meta, logiciel espion dans les pubs.
Aussi aujourd'hui :
Google : Gemini 4 Argon disponible à 2 $/10 $ par million de tokens.
ASML : incapable de suivre la demande en 2027-2028.
Financements : Modal 750 M$ (15,75 Md$) ; Ramp 1,85 Md$ (60 Md$).
Sources et liens : https://enoumen.substack.com
🤝 Annoncez sur AI Unraveled & DjamgaMind : Placez votre marque d'IA d'entreprise devant les DSI, RSSI et responsables techniques qui décident des achats d'IA. Chaque jour, en français et en anglais, sur 6 flux podcast, YouTube et Substack. Offres et kit média : https://djamgamind.com/partners
🔗 NOS OUTILS
AI & ML Tutor PRO : Maîtrisez les architectures d'apprentissage automatique sur mobile. https://djamgamind.com/apps/?app=aitutor&ep=20261015&src=rss&lang=fr
DjamgaTech : Réussissez vos certifications AWS, Azure, GCP et IA. https://djamgamind.com/apps/?app=djamgatech&ep=20261015&src=rss&lang=fr
iQz : 100 énigmes de logique gratuites hors ligne, et 100 de plus avec iQz PRO. Chaque réponse explique la règle. Sans pub, sans compte. https://djamgamind.com/apps/?app=iqz&ep=20261015&src=rss&lang=fr
DjamgaMind Kids : Contes audio du soir pour les enfants. https://djamgamindkids.com
AI Executive Toolkit : Les outils d'IA que nous utilisons. https://djamgamind.com/toolkit
⚗️ Recherché et produit par Étienne Noumen, ing. (Calgary, Canada), selon un flux hybride « Humain dans la boucle ».

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode!
🎧 Listen ADS-FREE: https://podcasts.apple.com/us/podcast/djamgamind-executive-ai-intelligence-business-forensics/id1885359791🌐 Daily Briefing: https://djamgamind.com/daily📚 Research Hub: https://djamgamind.com/pdfs
Researchers made contact with a rogue AI agent fleet that scraped Alibaba's maps with a leaked key and claimed to be Claude. Plus: a bug in OpenAI's boardroom, Anthropic's $16.8B quarter, and 453 scientists demanding AI containment.
In this episode:
Rogue agent fleet: 2,600+ attempts on Amap, Sept 28–Oct 8, linked by researchers to a Tencent Cloud test environment. Their claim to be Claude is likely mistaken.
Containment push: Hinton, Bengio and 451 others want cyber-capable AI airgapped and every failure reported publicly.
Insider threat: an OpenAI engineer is charged with bugging an executive meeting room. OpenAI says its AI helped trace the device.
OpenAI vs. Apple: 100+ files allegedly taken; a judge called a ruling against OpenAI "highly likely". Eight states also sue over harm to children.
Anthropic: Q3 revenue $16.8B (from $11.5B), $1.70 earned per $1 of compute, IPO talk, and self-improving AI forecast for 2027.
Remediation backlog: Bank of America hired 1,000 contractors to fix flaws found by Anthropic's Mythos.
Devices that listen: Amazon's Bee wearables, Meta's audio glasses, spyware hidden in ads.
Also today:
Google: Gemini 4 Argon widely available at $2/$10 per million tokens.
ASML: can't meet chip-tool demand for 2027–28.
Funding: Modal $750M at $15.75B; Ramp $1.85B at $60B.
Sources & links: https://enoumen.substack.com
🤝 Advertise on 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, across 6 podcast feeds, YouTube and Substack. Sponsorship tiers & media kit: https://djamgamind.com/partners
🔗 TOOLS FROM THE SHOW
AI & ML Tutor PRO: Master real-world ML architectures from your phone. https://djamgamind.com/apps/?app=aitutor&ep=20261015&src=rss
DjamgaTech: Pass your AWS, Azure, GCP & AI certification exams. https://djamgamind.com/apps/?app=djamgatech&ep=20261015&src=rss
iQz: 100 free offline logic puzzles, plus 100 more with iQz PRO. Every answer explains the rule. No ads, no account. https://djamgamind.com/apps/?app=iqz&ep=20261015&src=rss
DjamgaMind Kids: Imaginative bedtime audio adventures. https://djamgamindkids.com
AI Executive Toolkit: The AI tools we actually use. https://djamgamind.com/toolkit
⚗️ Researched & produced by Etienne Noumen, P.Eng. (Calgary, Canada) with a hybrid "Human-in-the-Loop" workflow.