What's really happening in AI.

Salesforce buys Fin (formerly Intercom) for $3.6 billion to bring proven AI agent technology and talent into its Agentforce offering.

• Purchase price: $3.6 billion
• Fin’s AI agent supports live chat, WhatsApp, SMS, phone calls, Slack and more
• Salesforce intends to use Fin’s team and technology to enhance Agentforce
• Deal expected to close in Salesforce’s fiscal 2027 final quarter

The deal accelerates Salesforce’s ability to offer enterprise AI agents for customer service across channels, which can reduce time‑to‑value for companies seeking automated support.

For you Review your CRM and support automation plans: evaluate whether tighter Salesforce AI agent capabilities could simplify your roadmap or require integration planning.

Sarvam raises $234 million in a Series B round, valuing the Bengaluru startup at $1.5 billion.

• So far $234M raised in the round; company aims to raise up to $300M in Series B.
• HCLTech is lead strategic investor with $150M committed; Bessemer, Khosla, Peak XV also participating.
• Sarvam previously raised about $41M in seed and Series A and released open‑source 30B and 105B parameter models earlier this year.
• Products focus on Indian languages and use cases; deployments reported in banking, insurance, government services and defense.

The funding plus a deep‑pocketed strategic partner gives Sarvam scale to commercialize a full‑stack AI offering—important for businesses seeking local language models, sovereign AI capabilities, or enterprise integrations in India.

For you Assess whether Indian‑localized models or a partnership with firms like Sarvam could serve your business needs or supply‑chain constraints; verify data‑sovereignty and compliance before integration.

Volunteer AR scans from Pokémon Go helped train Niantic Spatial models that are now combined with Vantor’s drone navigation software to enable GPS‑free positioning.

• Niantic added incentives in 2021 for players to scan real‑world locations; those scans trained Niantic Spatial’s foundation models.
• Niantic Spatial and defense firm Vantor pair Niantic’s ground‑level Visual Positioning System with Vantor’s Raptor software and 3D terrain data.
• Early tests report up to 70% error reduction and roughly 1.5‑meter accuracy; visual system resists standard jamming techniques.
• Niantic states the raw player scans were not handed directly to Vantor and that participation was voluntary and covered by privacy/ToS in place at the time.

This is a clear example of civilian crowdsourced mapping data moving into dual‑use applications, raising implications for privacy, public trust and how companies manage user‑generated spatial data.

For you Review terms and consent processes for any AR or crowdsourcing features you run; explicitly evaluate and disclose potential dual‑use or third‑party downstream uses of collected spatial data.

The main point: Cybersecurity startup NewCore exits stealth with $66M in seed funding to build an identity and governance platform for AI agents in the workplace.

• Seed round: $66M led by Cyberstarts, with Index Ventures and Evolution Equity Partners
• Post‑money valuation: $300M
• Platform purpose: manage human and AI‑agent identities together with agent‑specific permissions, lifecycle controls and revocation
• Founders: CEO Zohar Alon, CTO Amihai Neiderman, CRO Erez Yarkoni

As companies treat AI agents as workplace participants, traditional IAM systems may not scale or provide needed controls; a dedicated identity layer for agents addresses access, auditing and revocation challenges that raise security and compliance risk.

For you Takeaway: Audit your IAM for non‑human identities, map where agents will need unique permissions, and plan for lifecycle and revocation controls before deploying autonomous agents.

The main point: In April, Loft Orbital's satellite Yam‑9 ran DeepMind's Gemma 3 VLM onboard to autonomously identify areas of interest without ground analysts.

• First reported in‑orbit use of a vision‑language model; demonstration took place in April
• Platform: Yam‑9 by Loft Orbital; model: Gemma 3 from Google DeepMind, built for edge deployments
• Tasks performed: classify where natural environment meets human development; identify infrastructure around railway hubs
• Implication: on‑orbit triage reduces raw data downlink and enables always‑on patrol‑style monitoring

Onboard VLM processing can make space sensors more valuable by filtering and flagging data before downlink, speeding response and lowering ground‑processing costs — important for businesses buying or integrating satellite data.

For you Takeaway: Ask geo‑data providers whether they offer on‑satellite preprocessing or event‑based alerts — this can cut data transfer costs and speed up actionable insights.

The European Commission is assessing the practical impact after a US export control order led Anthropic to shut down Fable 5 and Mythos 5 for non‑US citizens.

• US government order prompted Anthropic to disable access to Fable 5 and Mythos 5 for users outside the US (per earlier reports).
• EU Commission says it is evaluating practical effects and warns emergency measures must not discriminate against partners.
• European researchers call the incident a wake‑up call, debating whether to boost resilience, develop local models, or invest jointly in chips and energy‑efficient computing.
• Experts cite risk: foreign government orders can cut off access overnight; proposals include coordinated European investment ('Airbus moment').

The episode highlights sovereign risk for companies relying on foreign AI providers — access to critical models can be disrupted by geopolitical or national‑security decisions.

For you Map which business processes depend on non‑EU third‑party models and prepare contingencies—contractual availability clauses, fallbacks, or local/european alternatives.

Microsoft CEO Satya Nadella says companies must build proprietary learning systems and retain 'token capital' or risk a small number of AI systems capturing most value.

• Nadella prescribes private evaluations, internal training on company data, and queryable, reusable institutional knowledge.
• Companies must be able to swap base models without losing built-up knowledge.
• Microsoft trains its own models and is tying customers to Azure and Office tooling while warning of value concentration.

Value in AI may shift from base models to firm‑specific learning loops and data ownership; firms that don't build these capabilities risk losing competitive and economic leverage to platform owners.

For you Assess which internal knowledge and data can become reusable training assets and plan how to integrate them into controlled learning loops or partnerships.

Tech companies increasingly cite AI as the reason for layoffs while data show layoffs accelerating in 2026.

• TrueUp reports about 363 layoff events so far in 2026, nearly 150,000 people affected — roughly 974 people per day and 44% faster than last year.
• Last month saw nearly 40,000 cuts, the highest single‑month total in two years.
• Outplacement firm Challenger, Grey & Christmas finds AI is the most‑cited reason across industries for three months running.
• Observers like Marc Andreessen and cases at Block and Uber highlight skepticism that AI is the primary cause versus pandemic‑era overhiring.

Labeling layoffs as driven by AI can mask other causes such as overstaffing and strategic restructuring; that affects labour markets, regulatory scrutiny, and how businesses communicate change.

For you If you manage or advise a SME, audit hiring levels and the real ROI of AI tools before using AI as a rationale for cuts; document automation impacts on headcount and reskilling needs.

Reports say the White House suspects a China‑linked group accessed Anthropic’s powerful model Mythos (and possibly Fable).

• Report by Semafor, summarized in The Verge
• Suspected access concerns Mythos 5 and/or Fable 5
• White House has not publicly confirmed China involvement; Anthropic declined to comment
• Officials fear access or distillation could enable reverse‑engineering and national security risks

Access by a foreign actor to cutting‑edge models raises risks of replication, misuse, or strategic exploitation; it helps explain recent export controls and matters for vendors, compliance, and risk planning in businesses.

For you Audit your AI vendors for export controls and model provenance, and require contractual guarantees about access controls and compliance with export regulations.

OpenAI launches the Partner Network and commits $150 million to help global partners accelerate enterprise AI adoption, deployment and transformation.

• Commitment amount: $150 million.
• Purpose: accelerate enterprise adoption, deployment and transformation of AI.
• Target partners: global service providers, system integrators and resellers.

For SMEs this expands the pool of certified implementation partners and brings more funding and technical resources into OpenAI’s ecosystem, which can speed deployments and lower implementation risk.

For you Check whether your current IT providers can join the network or whether certified partners in the program offer more relevant expertise for your AI initiatives.

Google Cloud introduces Open Knowledge Format (OKF) v0.1, a minimal spec that represents organizational knowledge as directories of Markdown files with YAML frontmatter.

• OKF requires a single YAML field "type"; other fields (title, description, resource, tags, timestamps) are optional
• Documents link via Markdown links to form a knowledge graph; bundles are editor‑readable and GitHub‑renderable
• Designed to be cloud‑ and agent‑framework‑agnostic so producers and consumers are decoupled
• Google ships reference implementations: a BigQuery enrichment agent, a static HTML visualizer, and sample bundles

A portable, human‑readable knowledge format reduces bespoke engineering to assemble agent context and enables interoperability across catalogs and tools — valuable for SMEs adopting agentic assistants or knowledge indexing.

For you Assess exporting key datasets and docs to OKF bundles and test an OKF‑based retrieval pipeline to speed up agent context building.

KPMG withdraws a report about AI adoption after several case studies in it were exposed as fabricated or false.

• Report title: “Redefining excellence in the age of agentic AI” (October 2025).
• Named organisations (UBS, UK NHS, Swiss Federal Railways, Transport for London) disputed the claims.
• Errors flagged by GPTZero and verified by the Financial Times; KPMG removed the report and opened an investigation.
• GPTZero highlights sloppy sourcing and ‘vibe citing’; EY previously withdrew a similarly flawed report.

Consulting reports are treated as authoritative and get reused by both people and AI systems; fabricated claims in such reports can seed widespread misinformation and bad business or technical decisions.

For you Demand primary sources and verification before basing procurement or strategy on third‑party AI reports.

A US government export-control directive leads Anthropic to disable Fable 5 and Mythos 5 worldwide after reported security concerns.

• Directive arrived Friday evening; Anthropic completely turned off access to Fable 5 and Mythos 5 to comply.
• Reports say Amazon CEO Andy Jassy and at least five other firms raised concerns with the government.
• Anthropic says it received only verbal evidence of a narrow, non‑universal jailbreak and notes similar capabilities exist in other models like GPT‑5.5.
• Security research allegedly showed Fable could be prompted into producing information usable for cyberattacks; experts disagree on whether this constitutes a true jailbreak.

Government-ordered removals of frontier models demonstrate that access to critical AI capabilities can be curtailed rapidly for national-security reasons, affecting availability, vendor risk and international deployments.

For you Check which frontier models your vendors rely on and plan contingencies for sudden access restrictions or regional blocks.

Anthropic halts access to its newest models Fable 5 and Mythos 5 for all foreign nationals following a U.S. government directive, sparking debate in India about dependence on foreign AI.

• Shutdown: Access to Fable 5 and Mythos 5 suspended for foreign users, including foreign employees
• Reported trigger: U.S. government directive after alleged security concerns reportedly raised by Amazon CEO Andy Jassy
• India link: Decision comes shortly after Anthropic partnered with Tata Consultancy Services to expand enterprise AI in India

The episode highlights how geopolitical actions can abruptly restrict access to frontier models — a direct operational risk for businesses that rely on a small set of external providers.

For you Assess your reliance on single-provider frontier models; plan fallback options such as open-source stacks, multi-provider architectures, or local suppliers for mission‑critical systems.

Meta cuts operational ties and halts data sharing with Manus as Beijing presses a divestiture order on national security grounds.

• Meta has implemented an operational separation of the roughly $2 billion acquisition of Manus, blocking employees from using Manus tools internally (Bloomberg).
• Manus had moved staff to Singapore earlier and is reportedly in talks to raise about $1 billion from outside investors to repurchase itself and pursue a Hong Kong listing via a Chinese joint‑venture structure.
• Chinese authorities are broadening controls—expanding travel restrictions for researchers/executives and requiring state sign‑off for some US investments in top AI firms.
• Manus continues shipping features, including integrations with Similarweb and Shopify, despite the separation.

This underlines how geopolitical and national‑security regulation can rapidly disrupt AI M&A, data access and product roadmaps; any company active in China or with Chinese targets must plan for forced divestiture and tighter capital controls.

For you Review contracts and data flows with China‑linked entities; build playbooks for rapid operational isolation and alternative investor scenarios in case of government‑driven divestiture.

Anthropic’s Claude Fable 5 posts leading scores on the FrontierMath benchmark, outperforming GPT‑5.5 by a notable margin.

• Epoch AI reports Fable 5 at 87% accuracy on tiers 1–3 and 88% on the hardest tier 4 (v2).
• OpenAI’s GPT‑5.5 scores about 75% on the same tier; earlier models were far lower.
• Results use FrontierMath’s standard scaffold with maximum reasoning effort.

Big improvements on hard math benchmarks imply better real‑world reasoning for tasks like modeling, optimization and technical QA — capabilities that matter for advanced automation.

For you Ask vendors for benchmark evidence in domains you care about and validate model reasoning on your own sample problems.