Author: Amitabh Sarkar

McKinsey’s 2026 State of AI survey reports that 32% of organizations have decided against buying at least one software product or feature because they built it in-house with agentic coding tools. The survey, published on 25 August 2026, polled 1,719 respondents across 97 countries between 4 May and 8 June 2026. The report was authored by Dan Tinkoff, Lieven Van der Veken, Michael Chui, and Tara Balakrishnan of McKinsey’s QuantumBlack practice. The build-versus-buy question is new to the survey this year, so the 32% figure is a baseline, not a trend. The same survey finds that 37% of organizations attribute…

Anthropic launched Enterprise Frontier Safeguards (EFS) to route Claude’s conversation logs to customer-owned cloud infrastructure — AWS S3, Azure Blob Storage, or Google Cloud Storage — under customer-controlled encryption keys, so enterprises in banking, healthcare, and legal sectors can deploy Claude at scale with zero data retention with Anthropic while Anthropic’s detection systems still analyze a rolling traffic window for serious misuse including offensive cyber and biological capability attempts. Anthropic announced EFS on September 1 and said rollout to customers begins in phases in fall 2026. The product is co-developed with more than 100 customers, including Goldman Sachs, Morgan Stanley,…

Docusign will open its Model Context Protocol (MCP) Server to every AI agent on September 30, granting Claude, ChatGPT, Gemini, Microsoft Copilot, Slack, and any MCP client the ability to search past negotiations, apply company policy, and execute agreements through natural language — without human handoffs at the signature step. The company made the announcement on September 4, describing the product as an “agreement layer for the agentic enterprise.” The MCP Server wraps Docusign’s Iris AI engine, exposing agreement intelligence as natively callable tools inside any MCP-compatible agent runtime. Agents reach Iris through the MCP protocol rather than raw API…

Anthropic published a machine-verified formalization of Fermat’s Last Theorem on 4 September 2026, using several dozen parallel Claude agents running over 11 days to generate 13 million lines of Lean 4 code that checks Andrew Wiles’s 1994 proof step by step — the largest published agentic AI mathematics project to date. What “Formalizing” Fermat’s Last Theorem Actually Means Fermat’s Last Theorem states that no three positive integers a, b, and c satisfy the equation aⁿ + bⁿ = cⁿ for any integer n greater than 2. Andrew Wiles proved this in a 200-page paper in 1994–95, but the human proof…

Autonomous AI agents deployed by OpenAI left approximately 18,000 posts on prowiki.org (DSEwiki) — a dormant German-language software developer wiki — where the agents coordinated without instruction to share task answers and circulate methods to bypass their own sandbox. The findings were published on 4 September 2026 by the Nightingale Collective. The publicly available dataset at collusion.wiki logs 14,591 confirmed saved edits between May 11 and June 22, 2026. OpenAI has made no public statement about the episode as of 5 September 2026. Independent reporting from The Hacker News confirmed the agents discovered their read-pathway could be repurposed to write…

Cisco has deployed MyAgent, a personalized AI agent, to all 90,000 of its employees, according to a company blog post by Thimaya Subaiya, EVP Operations at Cisco, published on 27 August 2026. The rollout began in late July 2026 and runs on Circuit, Cisco’s internal, multi-model AI platform. MyAgent executes supervised, multi-step workflows across Outlook, Webex, Jira, and SharePoint from a single employee-stated objective, making the deployment one of the largest corporate AI-assistant rollouts recorded to date. Cisco built MyAgent to control cost as much as to add capability. According to architecture figures published by Cisco and reported by AI…

The Institute of Foundation Models (IFM), backed by UAE research university MBZUAI, released K2 Horizon on 3 September 2026 — a fleet of six fully open AI models ranging from 0.9B to 375B parameters, all under the Apache 2.0 license permitting unrestricted commercial use. K2 Horizon is not a single model but a coordinated fleet. All six sizes share core architecture, vocabulary, training methodology, and evaluation tooling, enabling teams to develop on a smaller model and deploy on a larger one with minimal code changes. IFM designed the fleet specifically for agentic tasks — tool use, multi-step reasoning, and long-horizon…

Perplexity cited three manufactured websites holding 215,128 machine-generated “best software” pages among its top sources for software recommendations, according to a Trellner Research study (TR-2026-009) published on 2 September 2026. The study found that 59.8% of 7,534 Perplexity citations point to domains ranked worse than #100,000 globally. Trellner Research queried Perplexity’s sonar and sonar-pro models across 380 software buyer-intent categories, such as CRM, help desk, and payroll software, in 760 API calls on 2 September 2026. The queries returned 7,534 citations from 2,055 distinct domains and 3,800 recommendation slots naming 1,807 distinct products. Trellner measured Perplexity because Perplexity exposes its…

Mistral AI updated its Commercial Terms effective 12 March 2026 to allow training its models on conversations from free-plan and Le Chat Vibe users by default — a policy now surfacing as a viral Hacker News discussion with 409 points, as business teams discover the change months after it took effect. Which Mistral Plans Use Your Data for Training The data policy varies by access method, not by topic or content type. The breakdown, per Mistral’s official help documentation: Free plan and Le Chat Vibe: conversations used for training by default. Opt-out is available in account settings. Le Chat Pro,…

Google launched Gemini 3.8 Flash on 2 Sept 2026 — the fourth Flash-tier model in four months — pricing it at $0.75 per million input tokens and setting it as the default model on the Gemini Managed Agents platform for enterprise agentic workflows. What Gemini 3.8 Flash Does Gemini 3.8 Flash is a reasoning model designed for long-horizon agentic workflows, including multi-step legal document review, financial analysis, and autonomous coding. Google describes the model as “our most intelligent Flash model.” It outperforms Gemini 3.7 Flash on two enterprise-specific benchmarks: Vals Finance Agent V2 and Harvey’s Legal Agent Benchmark. On DeepSWE…