BankingNewsAI Daily Brief ·
Thinking Machines released Inkling, a 975B-parameter open-weight model enterprises can self-host.
Banking AI
Financial institutions & fintech technology
HSBC is helping set APAC rules for who is liable when payment agents act
HSBC has joined the Emerging Payments Association Asia as a founding member of its AI and Agentic Payments Working Group. The group brings banks, payment networks, fintechs and technology firms together to define standards for safe agentic commerce, with liability a central issue.
Action
Assign payments, legal and fraud leaders to define agent authority, authentication, transaction limits and loss allocation now; those controls are likely to become the market baseline in APAC.
Singapore now requires clear notices when firms use personal data to train AI
Singapore’s PDPC has made AI-specific notification mandatory for organizations using personal data to train AI models, alongside requirements around consent, safeguards and PDPA compliance. The change raises the bar from general privacy disclosures to explicit disclosure of AI-training uses.
Action
Inventory every GenAI training and retrieval workflow touching Singapore personal data, then update notices, consent paths and vendor terms before the new disclosure expectation becomes an examination issue.
General AI
Large language models & AI infrastructure
Thinking Machines released Inkling, a permissively licensed 975B-parameter multimodal open-weight model that enterprises can customize and self-host.
Thinking Machines Lab launched Inkling, its first model: a mixture-of-experts system with roughly 975B total parameters and 41B active parameters, accepting text, image, and audio inputs. The full weights are available under Apache 2.0, with up to 1M-token context in the open checkpoints and day-one availability across Hugging Face, Databricks, Modal, Baseten, vLLM, and SGLang. Independent testing cited in the newsletter ranks it as the leading U.S.-based open-weight release, though not ahead of the best closed models or leading Chinese open models on every evaluation.
Action
Benchmark Inkling now for controlled, self-hosted document intelligence and agent workflows where data residency, model customization, and vendor leverage matter; pressure your AI platform and core vendors to state when they will support it.
Alibaba has previewed a 2.4-trillion-parameter flagship model, intensifying open-model competition
Alibaba previewed Qwen3.8, its new flagship Qwen model with 2.4 trillion parameters, and claims it trails only Anthropic’s Claude Fable 5. The release extends the rapid rise of Chinese frontier-model alternatives and signals another potential option for enterprises seeking leverage against closed-model pricing and vendor concentration.
Action
Benchmark Qwen3.8 against incumbent models on your own regulated workloads, with particular focus on quality, data-residency options, security controls and total inference cost—not vendor benchmark claims.
Cognizant plans to put 100,000 employees on Gemini Enterprise
Cognizant is expanding its Google Cloud partnership with a plan to move 100,000 associates onto Gemini Enterprise. The scale makes this a meaningful test of whether governed enterprise AI can be embedded across a large services workforce rather than confined to technical pilots.
Action
Use Cognizant’s rollout as a benchmark for your own adoption plan: pair broad access with role-based use cases, approved data connections, training and measurable productivity targets.