The AI Brief

OpenAI introduced Presence on July 22 as a production deployment product for enterprise voice and chat agents, not a self-serve tool. The company is making it available through a limited general availability program; deployments are led by OpenAI Forward Deployed Engineers and selected systems integrators. OpenAI describes it as built through years of deploying agents with customers, shaped by the demands of mission-critical environments.

Presence supports voice and chat agents for customer support, outbound sales, and internal IT service requests; each deployment is scoped to a specific task, with the enterprise setting policies on what the agent can do, when it requires approval, and when a human should intervene. Presence powers OpenAI's own English-language phone support channel, handling open-ended requests, verifying callers, using account context, and taking approved actions; within weeks it met benchmarks used to grade frontline human support and now resolves 75% of inbound issues without human assistance, while a Codex-powered improvement loop reduced human handoffs by 15 percentage points in 10 days. Those figures are company-reported from OpenAI's own support line and have not been independently verified.

OpenAI named BBVA, SoftBank, and IAG among enterprises exploring or testing Presence-based voice support. The platform represents a strategic shift away from selling commoditized AI model access toward stickier enterprise software and consulting services. In May 2026, OpenAI launched the OpenAI Deployment Company, with investment and support from Bain & Company. Presence adds a software layer on top of that consulting infrastructure. OpenAI has not disclosed pricing or contractual terms; buyers should surface those early.

OpenAI launched Health in ChatGPT on July 23, 2026, rolling out to US users so they can securely connect their health information to better understand and navigate their health. Logged-in Free, Go, Plus, and Pro users in the United States who are 18 or older can connect supported health records and Apple Health data, see a dashboard of their health information, and ask questions grounded in their personal health context; Health lets users view lab results, medications, activity, sleep, and other health information in one place, and supports appointment preparation, test result interpretation, wellness tracking, and discussion preparation for healthcare professionals.

Every week, more than 300 million people turn to ChatGPT with health-related questions, but the context behind those questions is often scattered across patient portals, medical records, apps, and wearables. Health is built with layered privacy and security safeguards; connected medical records and Apple Health information, and conversations using them, are not used to train foundation models or target ads; and ChatGPT Health is designed to support, not replace, medical care and is not intended for diagnosis or treatment.

OpenAI is starting with a small group of early users, with Free, Go, Plus, and Pro plan holders outside the European Economic Area, Switzerland, and the United Kingdom eligible. OpenAI plans to expand access to all users on web and iOS in the coming weeks; medical record integrations and some apps are US-only at launch, and connecting Apple Health requires iOS.

The European Commission published guidelines on July 20, 2026 to help providers and deployers of AI systems meet the AI Act's transparency obligations, which start to apply on August 2, 2026. The publication gives every company that runs a chatbot, generates AI images, or publishes AI-written content for EU audiences 11 days to comply or face fines that can reach €15 million or 3 percent of worldwide annual turnover, whichever is higher.

Systems placed on the EU market on or after August 2 must comply from that date; all Article 50 obligations, chatbot disclosure, deepfake labeling, AI-generated public-interest text labeling, apply from August 2 without exception. Providers of generative AI systems already on the EU market before August 2 have until December 2, 2026 to bring their machine-readable marking capabilities into conformity with Article 50(2), under the EU Digital Omnibus provisional agreement of May 2026. That transitional window applies to the marking requirement only; deployer disclosure duties have no equivalent grace period.

Image, audio, and video deepfakes generated or manipulated before August 2, 2026 do not need to be marked or labeled retroactively. Automated AI labels applied by platforms do not relieve the operator of its own disclosure obligation; platform-side labels may complement, but cannot replace, the operator's own disclosure. The guidelines are non-binding interpretive guidance; only the Court of Justice of the EU can give authoritative interpretation of the AI Act. Compliance teams should act on the guidelines now rather than wait for judicial clarity.

DeepSeek announced the July 24, 2026 retirement of its legacy deepseek-chat and deepseek-reasoner API aliases in June, giving operators roughly a month to migrate. The aliases pointed to the now-superseded DeepSeek V3 and R1 model generations; the replacement endpoints are deepseek-v4-flash and deepseek-v4-pro, which correspond to the current V4 generation released earlier in 2026. The cutover is hard: the deadline on DeepSeek's API is July 24, when the legacy deepseek-chat and deepseek-reasoner aliases stop responding, with no announced extension.

The practical exposure is concentrated in integrations built during DeepSeek's rapid adoption surge in early 2026, when the alias-based endpoints were the canonical references in quickstart guides. Operators who routed DeepSeek calls through aggregator layers, OpenRouter, LiteLLM, or custom gateway proxies, need to verify whether the aggregator has updated its upstream mapping, not just that their own application code has changed. Gateway-level credential rotation and audit-trail confirmation are the open items most teams will have missed. First covered in Vol. I, No. 59 as the DeepSeek API alias retirement; today is the live enforcement date.

Anthropic launched the Anthropic Economic Index connector for Claude, letting anyone explore AI usage data directly in chat; users can ask questions about occupations, tasks, and trends, with answers grounded in the Index and source data available in Claude, addressing hard questions about which jobs will change, which tasks are being automated, and what it means for specific fields.

The Anthropic Economic Index draws on Claude usage patterns across a large sample of production interactions to estimate how AI touches different occupational tasks, distinguishing between AI augmenting a task and AI replacing it outright. The connector surfaces that data via natural-language queries, allowing users to ask about specific job categories, compare automation exposure across industries, or probe how use patterns have shifted over time. The dataset carries its own methodological limits: it reflects observed Claude usage, not a census of AI adoption across all platforms, and researchers using it for policy purposes should treat it as a large-scale signal rather than a representative sample of the full economy.

Disclosure: Claude, which generates this brief, is built by Anthropic.