AI News Roundup — 2026-02-28 (Enterprise + Product)
Anthropic hands MCP to Linux Foundation. Gemini agents ship on Galaxy S26 and Pixel 10. Outreach puts sales signals inside any AI. Salesforce ties GPT-5 into Agentforce 360. And IBM says AI-driven attacks are already here.
TL;DR: The Model Context Protocol (MCP) — Anthropic's standard for connecting AI agents to external tools — has been donated to the Linux Foundation, signaling that MCP is becoming neutral infrastructure rather than a vendor product. Meanwhile, Google and Samsung shipped the first production-grade Gemini agentic features on mobile hardware, and Outreach became the latest enterprise software vendor to launch an MCP Server, letting any external AI agent query its revenue data directly. The week closed with IBM's 2026 X-Force Threat Index confirming that AI-driven cyberattacks are accelerating — and that most enterprises aren't ready.
🔥 Top Stories
1. Anthropic Donates Model Context Protocol (MCP) to the Linux Foundation, Launches Agentic AI Foundation
What happened: Anthropic announced it is donating the Model Context Protocol (MCP) — the open standard for connecting AI agents to external data sources and tools — to the Linux Foundation. Alongside the donation, Anthropic established the Agentic AI Foundation (AAIF), a vendor-neutral body that will govern MCP and related open-source agentic AI infrastructure. The move transfers MCP's long-term stewardship away from Anthropic specifically and to a neutral foundation, following the pattern set by Kubernetes and OpenTelemetry. MCP is already adopted by OpenAI, Google DeepMind, Microsoft, and hundreds of third-party integrators.
Why it matters: This is a significant strategic signal: Anthropic is betting that MCP wins by becoming a standard, not by staying proprietary. Donating to Linux Foundation means MCP is now governed like TCP/IP — not like a product. For enterprise architects and product teams, this increases the confidence case for building MCP-based integrations: the spec won't be controlled by any single AI lab. It also accelerates the pace at which enterprise software vendors (like Outreach — see below) will ship MCP Servers as table stakes. If you're designing AI integration architecture in 2026, MCP is now the de facto answer for interoperability.
Source: Anthropic blog · Yahoo Finance / Business Wire
2. Google and Samsung Deploy Gemini Agentic Features on Galaxy S26 and Pixel 10
What happened: At Galaxy Unpacked 2026 in San Francisco on February 25, Samsung Electronics unveiled the Galaxy S26, S26 Plus, and S26 Ultra — and announced a multi-engine AI architecture: Google Gemini for agentic cross-app tasks (booking rides, sending messages, ordering food without manual navigation), Perplexity AI for web search queries, and an upgraded Samsung Bixby as the on-device assistant. Google simultaneously previewed new Gemini agentic capabilities coming to both Galaxy S26 and Pixel 10 — executing multi-step tasks inside apps using a secure virtual window, with scam call detection as a launch feature alongside task orchestration.
Why it matters: This is the first mass-market hardware deployment of multi-step AI agents in a consumer device — and it's shipping before Apple's redesigned Siri debuts. The Galaxy S26's three-engine model (Gemini for tasks, Perplexity for search, Bixby on-device) signals a future where mobile AI is composable rather than monolithic. For enterprise product teams: mobile agents that can act across apps set a new bar for what employees expect from internal tools. The "agent executes a multi-step task inside a secure virtual window" interaction model will migrate into enterprise software — plan your UX strategy accordingly.
Source: Samsung Newsroom · CNBC · gHacks
3. Salesforce Agentforce 360 Integrates GPT-5 and Deepens OpenAI + Google Ties
What happened: Salesforce announced expanded integrations connecting its Agentforce 360 platform more deeply with both OpenAI and Google. Through the OpenAI partnership, Salesforce's Agentforce 360 (including sales records, customer conversation data, and Tableau visualizations) becomes accessible via natural language inside ChatGPT. Enterprise customers can also use OpenAI's GPT-5 — described as a unified model that selects between fast response and extended reasoning modes — to build AI agents and prompts inside the Salesforce Platform. The Google integration adds Gemini models as an option within Salesforce's AI Studio.
Why it matters: Salesforce is threading a needle: rather than building a proprietary model stack, it's becoming the enterprise data layer that connects to whichever frontier model a customer prefers — OpenAI, Google, or Anthropic. GPT-5 availability inside the Salesforce Platform means enterprises can run state-of-the-art reasoning on their CRM data without moving it out of Salesforce's trust layer. For teams using Salesforce: the practical implication is that your sales reps may soon build workflow agents directly in Salesforce Studio, powered by GPT-5, with no separate API contract needed.
4. Gartner: 40% of Enterprise Applications Will Include AI Agents by End of 2026
What happened: Gartner published its latest enterprise AI adoption forecast, projecting that by end of 2026 approximately 40% of enterprise applications will integrate task-specific AI agents — up from under 5% in 2025. The forecast cites the rapid commercialization of agent platforms from OpenAI (Frontier), Anthropic (Cowork), Google (Vertex AI Agent Builder), Microsoft (Copilot Studio), and Salesforce (Agentforce) as the primary driver. Gartner also notes that agent management — identity, access control, observability — is emerging as a distinct infrastructure category.
Why it matters: A jump from 5% to 40% in one year is a seismic adoption curve — if it materializes. Gartner's number is aspirational, but the trajectory is real: every major enterprise platform is shipping agent capabilities right now. For product managers and IT leaders: this forecast validates treating agent integration as a 2026 roadmap priority, not a 2027 exploration. The corollary concern is the one Gartner flags — agent management infrastructure (who authorized this agent? what did it access?) is not keeping pace with deployment speed.
Source: Windows Forum / Gartner analysis
🚢 Shipping & Platform Updates
5. Outreach Launches MCP Server — GA, Revenue Signals Available to Any External AI Agent
What happened: Outreach, the agentic AI platform for revenue teams, announced general availability of its Model Context Protocol (MCP) Server on February 24, 2026. The Outreach MCP Server allows external AI systems — including Claude, GPT-5, and Gemini — to read Outreach's buyer intent signals, sales activity data, and next-best-action recommendations without requiring custom integration work or a dedicated API contract. The Server joins Anthropic's growing MCP ecosystem and ships alongside Outreach's February 2026 product release.
Why it matters: MCP Servers are rapidly becoming the enterprise integration layer that REST APIs were in 2012 — but optimized for AI agents rather than human developers. Outreach's MCP Server means a sales team's AI assistant (whatever model powers it) can pull deal risk signals and suggested actions from Outreach in real time, with no custom plumbing. For revenue operations and sales engineering teams: evaluate MCP Server support as a vendor selection criterion going forward. Tools that offer MCP Servers will integrate into your AI stack with dramatically less engineering effort than those that don't.
Source: Business Wire · Outreach blog
6. NTT DATA and AWS Announce Strategic Collaboration to Scale Agentic AI
What happened: NTT DATA and Amazon Web Services (AWS) announced a strategic collaboration focused on scaling enterprise agentic AI deployments. The partnership targets modernization of legacy systems and development of industry-specific cloud platforms — with agentic AI as the architectural centerpiece. NTT DATA will offer AWS-powered agentic transformation services to its enterprise client base across financial services, healthcare, and manufacturing sectors.
Why it matters: NTT DATA operates at the intersection of legacy enterprise infrastructure and modern cloud — exactly where agentic AI is hardest to deploy. A partnership with AWS that specifically targets legacy modernization suggests the market is moving from "greenfield agent experiments" to "retrofit existing systems." For enterprise leaders managing older tech stacks: this signals that agentic AI is being productized for your environment, not just for cloud-native startups. AWS Bedrock (the underlying model and agent service) continues to accumulate enterprise system integrator partnerships at a rapid pace.
Source: Forbes Enterprise AI
7. Reddit Tests AI-Powered Shopping Carousels Inside Native Search Results
What happened: Reddit is piloting interactive product carousels embedded within its native search results for select U.S. users, initially targeting consumer electronics queries. The carousels combine community-generated product recommendations with real-time product data from Dynamic Product Ads (DPA) partner catalogs — displaying pricing, images, and direct purchase links without leaving Reddit. The test is a direct bid to convert Reddit's high-intent search traffic into a retail media surface.
Why it matters: Reddit's user base approaches product decisions via community trust — "what does this community actually recommend?" — which is a fundamentally different (and often higher-trust) signal than algorithmic search results. Embedding DPA catalogs into that context creates a powerful commercial surface: social proof + product availability + purchase path in one interaction. For brands with DPA feeds already running on Meta or Google: explore Reddit DPA eligibility now. If the pilot scales, early catalog participants will have inventory advantage before it becomes competitive.
Source: PPC.land
8. Anthropic Acquires AI Model Evaluation Startup (Details Emerging)
What happened: Reporting from MarketingProfs and other outlets indicates Anthropic made an acquisition in the AI model evaluation space in late February 2026. Full acquisition details — name of acquired company, deal size, and specific capabilities — had not been fully confirmed by publication deadline. The acquisition is expected to strengthen Anthropic's enterprise reliability tooling for Claude deployments.
Why it matters: Model evaluation is the unsexy but critical piece of enterprise AI deployment — it's how you know whether your agent is behaving correctly before and after you ship it. If Anthropic is acquiring in this space, it signals that eval tooling will become a built-in Claude Enterprise capability rather than a third-party integration. For teams currently using external eval tools (Braintrust, LangSmith, etc.): worth tracking whether Anthropic's native offering closes the gap on your current stack.
Source: MarketingProfs AI Update
⚖️ Policy, Security & Governance
9. IBM 2026 X-Force Threat Index: AI-Driven Attacks Escalating, Basic Security Gaps Persist
What happened: IBM released its 2026 X-Force Threat Index, reporting that AI-driven cyberattacks are escalating across enterprise environments. Key findings: 13% of organizations reported breaches of AI models or AI-powered applications; 97% of organizations that reported AI system breaches also reported lacking proper AI access controls. The report flags prompt injection, model poisoning, and credential theft via AI-generated phishing as the top three AI-specific attack vectors. IBM X-Force also notes that attackers are increasingly using AI agents to automate reconnaissance and lateral movement inside enterprise networks.
Why it matters: The 97% figure is alarming: almost every organization that suffered an AI breach did so because it hadn't implemented basic access governance for its AI systems. As enterprises rush to deploy AI agents with broad system access, the attack surface expands dramatically. For security teams: AI access controls are not optional overhead — they're the minimum bar. Start with the basics: least-privilege access for AI agents, audit logging on all agent actions, and prompt injection testing for any AI system that reads external data. The SC World prediction that a major breach will trace back to an AI agent in 2026 looks increasingly credible.
Source: IBM Newsroom · SC World
10. Colorado AI Act: Legislature Weighing Amendments Before June 30 Deadline
What happened: Colorado's General Assembly is actively considering amendments to the Colorado Artificial Intelligence Act (SB 24-205) during its 2026 session. The law — which requires impact assessments, consumer transparency disclosures, and anti-discrimination protections for "high-risk" AI systems — is on a June 30, 2026 compliance deadline following a prior delay from February 1 (signed by Governor Jared Polis via SB 25B-004). Proposed amendments would clarify scope definitions and adjust developer versus deployer liability boundaries. No amendment has passed as of February 28.
Why it matters: Don't wait for amendments to start compliance work. The most likely outcome is that the June 30 date holds and the law's core requirements remain: impact assessments for high-risk AI use cases (hiring, credit, housing, healthcare, education), consumer disclosures, and documentation of AI decision-making. Legal teams at companies operating in Colorado or serving Colorado residents should have a risk inventory of AI systems underway now. Four months is not a long runway for a compliance program that requires internal documentation, vendor assessments, and policy updates.
Source: Greenberg Traurig · Colorado General Assembly · Drata AI Compliance Guide
💡 One Take: Open Infrastructure Wins
The most strategically significant move this week wasn't a product launch — it was Anthropic donating MCP to the Linux Foundation. On the surface, it looks like a governance announcement. But the subtext is: Anthropic is betting that the fastest path to winning enterprise AI is to become the architect of the standard, not the gatekeeper of the protocol.
The playbook is familiar. Google donated Kubernetes to the CNCF and then became the dominant Kubernetes cloud vendor. The Linux Foundation itself is home to the tooling that powers nearly every enterprise system. By giving MCP neutral governance while maintaining deep investment in the ecosystem (Claude as the best MCP-native agent, Anthropic as a founding AAIF member), Anthropic gets the enterprise trust benefits of open standards without ceding commercial position.
The downstream effect for enterprise buyers: if MCP becomes the universal interface for AI agent-to-tool connectivity — and it's trending that way rapidly — then your AI integration investments become more portable, less vendor-locked, and more auditable. That's a good outcome for enterprise architects. Start designing for MCP now, before every vendor has a proprietary alternative they're trying to lock you into.
What to do this week:
- If you're designing AI agent architecture: MCP just became the de facto interoperability standard with Linux Foundation governance. Evaluate MCP Server availability as a procurement requirement for any new AI-adjacent software vendor.
- If you use Outreach: The new MCP Server means your external AI assistant can query deal data directly. Test it with Claude or GPT-5 before building any custom CRM integration — you may not need one.
- If you're a security leader: Read the IBM X-Force 2026 findings and answer honestly: do you have access controls specifically governing your AI agents? If not, that's a P0 gap given current threat trends.
- If your company deploys AI in hiring, credit, or healthcare in Colorado: The June 30 compliance deadline for the Colorado AI Act is four months away. Assign an owner and begin the impact assessment process this week — don't assume amendments will extend the deadline again.