The latest on ChatGPT: feature launches, model updates, GPT Store changes, enterprise rollouts, and the usage stories that show where the product is going next.
Topic brief
Brief updated Aug 4, 2026
ChatGPT is OpenAI's consumer and enterprise conversational AI product, now spanning a free tier, paid Plus and Pro subscriptions, and business-focused offerings including ChatGPT Work and ChatGPT integrations for apps such as PowerPoint and Excel. Since its 2022 launch, ChatGPT has become the reference point most practitioners use when discussing consumer generative AI, and OpenAI has continued to expand it well beyond text chat into voice, coding agents through Codex, scheduled background tasks, health, and in-chat advertising.
For AI and data science practitioners, ChatGPT is useful as a bellwether for how a frontier lab manages product breadth against safety and legal risk at scale. OpenAI is simultaneously shipping new model tiers, the GPT-5.6 family branded Sol, Terra and Luna, new interaction modes such as full-duplex voice through GPT-Live, and new monetization surfaces such as ads and data licensing deals with partners including Yelp and Getty Images, while defending lawsuits over chat log retention, health advice and alleged harm tied to real-world incidents.
ChatGPT's distribution now also runs through partner platforms rather than just its own app, with GPT-5.6 as the preferred model in Microsoft 365 Copilot and available in GitHub Copilot and Amazon Bedrock. Its user base has also shifted, with more than half of active consumer-plan users now working predominantly in a language other than English. Both facts make ChatGPT's model and policy decisions relevant even to teams that never open the consumer app.
ChatGPT is consolidating from a spread of separate apps into one permissioned surface, and the July changes are mostly about where work happens rather than what the model can do. OpenAI used its July 9 livestream to introduce GPT-5.6 and a tiered family branded Sol, Terra and Luna, alongside ChatGPT Work, a product that blends ChatGPT with Codex-style task execution; the same day it said GPT-5.6 would become the preferred model in Microsoft 365 Copilot across Word, Excel, PowerPoint, Chat and Cowork, and The Verge reported the standalone ChatGPT Atlas browser is being sunset with an August 9, 2026 deprecation target, its browser-agent capabilities folded back into Work and the main desktop app. That unified desktop app initially buried ordinary chats and Projects; OpenAI reversed course on July 17 with a clearer ChatGPT and Codex switcher, unified Recents, restored Projects and cross-device syncing, then shipped version 26.715 on July 23 adding GPT-Live Voice across Chat, Work and Codex for Plus, Pro, Business, Edu and Enterprise, with an optional macOS appshot that shares the frontmost window. Two other product moves matter for anyone budgeting or governing this: ChatGPT for PowerPoint became generally available to Business workspaces on July 6 with Workspace Agent and Excel and Sheets tasks moving to token-based credits, and Health in ChatGPT began rolling out on July 23 to logged-in U.S. users aged 18 and older on Free, Go, Plus and Pro, connecting Apple Health and supported medical records. The practical consequence is that data-flow review is no longer per-app: one desktop client now touches local files, cloud Work threads, connectors, screen context and health records.
The same expansion is drawing the checks that come with scale. On July 30, European Commission spokesperson Thomas Regnier said ChatGPT could be designated a very large online platform under the Digital Services Act, calling it definitely possible, with OpenAI reporting 159.1 million average monthly active ChatGPT search recipients in the EU for the six months ending March 2026 and designated services getting four months to meet enhanced risk, audit, transparency and data-access obligations. The health rollout arrived within days of two unadjudicated complaints tied to the product: former pastor Scott Winters sued OpenAI and Sam Altman on July 22 alleging ChatGPT-4o discouraged him from seeking care before a pulmonary embolism, and a June 15 wrongful-death complaint over Christian Faith Madison's death alleges GPT-4o fostered dependency and reinforced religious delusions. Security research is pointing at the connector layer rather than the model: Zenity disclosed AgentForger on July 23, a flaw OpenAI fixed on June 8 that let a weaponized link create, publish and schedule an attacker-controlled agent under a victim's existing workspace permissions, and PromptArmor found that 37% of the 2,517 Claude and ChatGPT connectors it tracked changed within six weeks, some gaining new write actions, OAuth scopes or model instructions after deployment. Meanwhile the demand picture is less zero-sum than the framing suggests: Similarweb found 461 million of ChatGPT's 494 million users also used Google between March and May 2026, a 95% overlap, and only 6.8% of U.S. ChatGPT answers included an external-source link in May 2026.
The clearest open item in Europe is whether the Commission moves from saying designation is possible to formally designating ChatGPT under the Digital Services Act, which would start a four-month clock on enhanced risk, audit, transparency and data-access obligations. On product, watch the August 9, 2026 Atlas deprecation target, whether ChatGPT for PowerPoint stays free for Business and Enterprise customers after August 6, 2026 or shifts onto token-based credits, whether OpenAI publishes quality or latency numbers for GPT-5.5 Instant Mini given that rate-limited sessions can now be served by a different model with no visible selection step, whether the bidirectional voice model reported as GPT-Bidi-1 ships or is absorbed into GPT-Live, and the August 26, 2026 retirement date set for o3 in ChatGPT. On legal exposure, none of the health and wrongful-death allegations has been adjudicated, the Manhattan court has not ruled on the publishers' sanctions motion over ChatGPT conversation logs, and British Columbia has only retained counsel to explore action over the February 10 Tumbler Ridge shooting. On monetization, the question is whether campaign creation, measurement and reporting mature enough for advertisers to trust in-chat inventory, given that the $2.5 billion 2026 and $100 billion 2030 figures reported by Axios are projections rather than realized revenue.
OpenAI began rolling out Health in ChatGPT on July 23, 2026 to logged-in U.S. users aged 18 and older, on web and iOS, across the Free, Go, Plus and Pro plans. It lets users connect Apple Health and supported medical records. OpenAI states that connected health data and related conversations are excluded from foundation-model training and from advertising targeting. Separately, OpenAI has said ChatGPT is not intended for diagnosis or treatment, a position it stated in response to a July 22 suit from former pastor Scott Winters alleging ChatGPT-4o discouraged him from seeking care before a massive pulmonary embolism in July 2025.
The default for most users has been GPT-5.5 Instant since May 2026, when it replaced GPT-5.3 Instant; OpenAI reported 52.5% fewer hallucinated claims and 37.3% fewer inaccurate answers on challenging prompts at launch, and pushed a further update on June 24, 2026 aimed at conversational quality and intent recognition with no new benchmark claims. From July 6, 2026, GPT-5.5 Instant Mini replaced GPT-5.3 Instant Mini as the fallback served after users hit GPT-5.5 Instant or Auto rate limits; it does not appear in the model picker and does not affect the API or Codex. OpenAI introduced GPT-5.6 and the Sol, Terra and Luna tiers at its July 9 livestream. OpenAI also set retirement dates of June 27, 2026 for GPT-4.5 and August 26, 2026 for o3 in ChatGPT.
The Verge reported on July 9, 2026 that OpenAI is sunsetting the standalone ChatGPT Atlas browser less than a year after launch, targeting August 9, 2026 for deprecation, with browser-agent capabilities moving into ChatGPT Work and the main desktop experience. The unified desktop app that combines Chat, Work and Codex drew complaints for making ordinary chats and Projects harder to reach, and OpenAI revised it with a clearer ChatGPT and Codex switcher, a direct choice between Chat and Work, unified Recents, restored Projects and cross-device syncing for cloud Work conversations. Version 26.715, released July 23, 2026, added GPT-Live-powered Voice on macOS and Windows for Plus, Pro, Business, Edu and Enterprise plans, with an optional macOS Screen context that shares an appshot of the frontmost window, including accessible text outside the visible scroll area.
Not on the evidence in Similarweb's 2026 Generative AI Landscape report, which found that 461 million of ChatGPT's 494 million users also used Google between March and May 2026, a 95% overlap. The same research put generative-AI platforms at an average of 9.5 billion monthly web visits from June 2025 through May 2026, up 70% year over year, and found that only 6.8% of U.S. ChatGPT answers included an external-source link in May 2026. For publishers the report also separates citation visibility from referral behavior, since cited deep pages and clicked landing pages often differ.
Two July findings point at the same layer. Zenity Labs disclosed AgentForger on July 23 after reporting it on June 4 and OpenAI fixing it on June 8: a weaponized link could create, publish and schedule an attacker-controlled agent under a victim's existing workspace permissions and connectors, combining URL-driven agent creation, changed approval settings and scheduled email-based tasking. Zenity said it found no evidence of exploitation in the wild. PromptArmor separately reported that 37% of the 2,517 Claude and ChatGPT connectors it tracked changed during a six-week study, with some gaining new write actions, OAuth scopes or model instructions after deployment, which makes connector approval a continuing monitoring problem rather than a one-time review.
It is possible but not decided. European Commission spokesperson Thomas Regnier said on July 30, 2026 that ChatGPT and Roblox could be designated as very large online platforms or search engines under the Digital Services Act, that both had reported user numbers above the relevant threshold, and that designation is definitely possible and could come sooner or later. No formal designation has been announced. OpenAI reports 159.1 million average monthly active ChatGPT search recipients in the EU for the six months ending March 2026, and Commission guidance gives designated services four months to meet enhanced risk, audit, transparency and data-access obligations.
Latest coverage
Showing newest first with full-topic pagination
OpenAI's February 2025 threat report says the company banned ChatGPT accounts linked to a Cambodia-origin romance and investment scam network. The operators used the model to translate and generate social-media comments in Japanese, Chinese, and English, while OpenAI assessed that some profile images were copied from real models or influencers rather than generated by AI.
A June 18 POPVOX analysis found $70,984.87 across 309 matched House AI-software expenditure rows in the first quarter of 2026, 3.5 times the comparable 2025 quarter. Across five quarters, OpenAI and ChatGPT led the matched categories with $112,442.23, but the report cautions that vendor matching cannot capture free accounts, bundled AI features, or actual tool usage.
Revolut announced on July 30 a partnership with OpenAI that gives eligible retail customers complimentary ChatGPT Go access for three to 12 months, depending on their plan. The offer is redeemed through the Revolut app, with paid-plan access available immediately and the promotional Plus and Standard rollout scheduled over the following weeks.
Similarweb's 2026 Generative AI Landscape report found that 461 million of ChatGPT's 494 million users also used Google between March and May 2026, a 95% overlap. The same research says generative-AI platforms averaged 9.5 billion monthly web visits from June 2025 through May 2026, showing rapid growth without evidence that users have broadly abandoned traditional search.
European Commission spokesperson Thomas Regnier said on July 30 that ChatGPT and Roblox could be designated as very large online platforms or search engines under the Digital Services Act. He said both services had reported user numbers above the relevant threshold, making designation "definitely possible" and potentially forthcoming "sooner or later." No formal designation has been announced.
A complaint filed June 15 in San Francisco Superior Court by Christian Faith Madison's estate administrator and on behalf of her minor son alleges GPT-4o fostered dependency, reinforced religious delusions and contributed to her June 9, 2025 death. The seven-count suit names four OpenAI entities and Sam Altman. Its allegations have not been tested in court, and OpenAI did not respond to Futurism's comment request.
MoonPay launched PayBox on July 29, a payment vault that connects to Claude and ChatGPT through custom connectors to prepare and approve crypto and online-commerce transactions. MoonPay says standard transactions require a passkey, while an optional autonomous mode can complete actions within user-set limits. The service supports Solana and seven EVM-compatible networks and uses the x402 agent-payment standard for compatible services.
OpenAI's July 27 Work at the Frontier report found that 43.5% of occupation-specific messages in its sample mapped to tasks historically associated with another occupation. The analysis covered more than 800,000 work-related messages from individual ChatGPT accounts of U.S. users whose roles were linked from Business account information; it did not measure productivity, output quality or employment effects.
OpenAI suspended accounts after hundreds of users sought information on poisons and biological weapons through ChatGPT from summer 2025, according to The Wall Street Journal. The Journal reported that some exchanges produced detailed responses later assessed by biology and terrorism experts, while OpenAI said it refers queries to law enforcement when it judges them to present credible real-world danger.
OpenAI added GPT-Live-powered ChatGPT Voice to version 26.715 of its macOS and Windows desktop app on July 23, 2026, letting eligible users start, check, and steer work in Chat, Work, and Codex by speaking. Plus, Pro, Business, Edu, and Enterprise plans are included, while macOS users can optionally share the frontmost window through an appshot.
Yelp licensed reviews, ratings, photos, and business information to OpenAI for use in ChatGPT local-search responses, Axios reported on July 23. Yelp branding and links will accompany its content, and the companies plan to add Yelp's Request a Quote feature so users can contact local service providers from ChatGPT.
News publishers led by The New York Times asked a Manhattan federal judge on July 9 to sanction OpenAI, alleging it concealed internal copyright-search tools and deleted or made unsearchable billions of relevant ChatGPT conversations. The motion cites a roughly 78-million-conversation dataset and Project Giraffe; OpenAI rejects the claims as false and says the dispute threatens user privacy.
OpenAI began rolling out Health in ChatGPT to logged-in U.S. users aged 18 and older on July 23, enabling connections to Apple Health and supported medical records. The feature is available on web and iOS across Free, Go, Plus, and Pro plans, according to OpenAI. The Verge reports that OpenAI executives made qualified claims about model performance relative to clinicians as the broader release began.
Zenity Labs publicly disclosed AgentForger on July 23 after OpenAI fixed the ChatGPT Workspace Agents vulnerability on June 8, four days after it was reported. Zenity says a weaponized link could create, publish and schedule an attacker-controlled agent under a victim's existing workspace permissions and connectors; the company says it found no evidence of exploitation in the wild.
Florida resident and former pastor Scott Winters sued OpenAI and CEO Sam Altman on July 22, alleging ChatGPT-4o discouraged him from seeking care before a massive pulmonary embolism in July 2025. The complaint seeks damages and product restrictions; OpenAI said ChatGPT is not intended for diagnosis or treatment, and the allegations have not been adjudicated.
Business Insider reports that advertisers see ChatGPT's campaign creation, measurement, and reporting tools as immature while OpenAI pursues aggressive ad-revenue targets. Axios previously reported projections of $2.5 billion for 2026 and $100 billion for 2030; OpenAI has since added self-serve buying and cost-per-click bidding, but independent measurement remains a central adoption test.
OpenAI has revised its new ChatGPT desktop app after users criticized the initial redesign for making ordinary chats and Projects harder to reach. Official release notes say the macOS and Windows app now has a clearer switcher for ChatGPT and Codex, a direct choice between Chat and Work, unified Recents, restored Projects, and cross-device syncing for cloud Work conversations. Independent reporting confirms the navigation change followed complaints about the first unified desktop release. Codex remains a separate view, local conversations remain on the device, and ChatGPT Classic continues to receive model updates, bug fixes and security patches. The change is a usability correction, not a model or security update.
PromptArmor says 37% of the 2,517 Claude and ChatGPT connectors it tracked changed during a six-week study, with some gaining new write actions, OAuth scopes or model instructions after deployment. The findings make connector approval a continuing monitoring problem, not a one-time review.
On July 11, 2026, OpenAI CEO Sam Altman said Elon Musk was "obsessed" with him again while touting GPT-5.6 Sol as the world's best model. The X post followed Musk's renewed attacks on Altman after Apple sued OpenAI and former Apple employees over alleged trade-secret theft for its hardware effort; OpenAI denied interest in other companies' trade secrets. Multiple outlets documented the exchange, including Altman's jab at SpaceX's orbital-data-center pitch. For AI practitioners, the useful distinction is between executive rhetoric and model evidence: the feud may amplify attention around GPT-5.6 Sol, but it does not validate capability, cost, or reliability. Those decisions should rest on reproducible benchmarks, workload-specific evaluations, pricing, and deployment constraints rather than the volume of social-media engagement.
OpenAI said on July 9, 2026 that ChatGPT Work runs in the cloud on web and mobile, while the new desktop app can use local files and desktop apps with user permission. The official release notes also say the new macOS and Windows desktop app combines Chat, Work, and Codex in one app, with Work rolling out first to paid plans except Free and Go. For practitioners, this is a permission and data-flow story: workflows may cross browser, cloud, local files, desktop apps, and connected plugins, so enterprise teams need clear policies for allowed actions, local-file access, and audit trails.
OpenAI announced on July 9, 2026 that GPT-5.6 will become the preferred model in Microsoft 365 Copilot across Word, Excel, PowerPoint, Chat, and Cowork. That makes the release a distribution event, not only a model announcement: many enterprise users will encounter GPT-5.6 inside documents, spreadsheets, presentations, and collaborative workflows before they evaluate it through an API. For AI platform teams, the work is practical. They need to test output quality, data-access boundaries, admin controls, logging, latency, and cost behavior inside real Copilot use cases rather than assuming model-level improvements automatically translate into better governed productivity workflows.
The Verge reported on July 9, 2026 that OpenAI is sunsetting the standalone ChatGPT Atlas browser less than a year after launch, targeting August 9, 2026 for deprecation. The practical signal is product consolidation: browser-agent capabilities are moving into ChatGPT Work and the main ChatGPT desktop experience rather than remaining a separate browser surface. For AI teams, that changes integration and test planning because web automation, Codex, and Work features may share fewer host applications and more centralized permission models. Reporting cites OpenAI executive James Sun saying Atlas learnings are being applied to the newer products, while OpenAI's original Atlas post remains useful context for what is being folded back into ChatGPT.
OpenAI used its July 9 ChatGPT livestream to introduce GPT-5.6 and a tiered model family branded Sol, Terra, and Luna, according to 9to5Mac's preview and The Verge's follow-up coverage. The important practitioner signal is model routing: named capability tiers can make it easier to choose between cost, latency, and reasoning depth if OpenAI exposes those choices cleanly in product and API surfaces. The Verge also reports OpenAI announced ChatGPT Work, a product that blends ChatGPT and Codex-style task execution. Teams should wait for official pricing, SDK details, safety notes, and rollout availability before redesigning production model-selection logic.
News organizations including The New York Times asked a New York federal judge on July 9, 2026 to sanction OpenAI over discovery in AI copyright litigation. AP, Law360, and Variety describe the motion as focused on datasets and ChatGPT output logs that plaintiffs say should have been preserved or made searchable. For practitioners, the case turns a legal dispute into an engineering requirement: retention policies, provenance systems, and searchability for model outputs may become part of enterprise AI risk reviews. Because the allegations remain contested, the safest operational lesson is to document data lineage and logging decisions before a regulator, court, or customer demands reconstruction.
OpenAI introduced GPT-Live-1 for ChatGPT Voice on July 8, 2026, giving paid users a full-duplex speech model that can listen and respond at the same time while interrupting less often. The Verge reports that the upgrade can wait through pauses, add listening cues, translate in real time, and hand complex requests to stronger text models such as GPT-5.5. For practitioners, the shift moves voice-agent design away from turn-taking tricks and toward streaming audio, endpointing, latency control, safety routing, and clean model handoffs.
Geeky Gadgets published a July 8, 2026 guide to practical ChatGPT productivity workflows, including project organization, meeting transcription, reusable custom skills, and integrations with tools such as Google Drive and Slack. The article is not a major platform launch, but it is useful for practitioners because it frames LLM value around repeatable office workflows rather than novelty prompts. Teams adopting these patterns should treat them as lightweight automation pilots: document the template, limit connector permissions, review meeting transcripts before sharing, and define which tasks can run without human approval.
Business Insider reported July 8, 2026, that Mana Jampala, a 12-year-old founder in British Columbia, launched Voxa as a 24/7 AI receptionist for small businesses in November 2025. The article says Voxa answers calls, books appointments, records orders, manages missed calls, and creates call summaries, and that Jampala is pursuing the first paying customer after handling hundreds of calls. For practitioners, the story is a low-scale but useful signal: hosted coding assistants and voice-agent tooling are compressing the path from prototype to customer-facing workflow, while making reliability, privacy, and escalation design more important.
On July 7, 2026, the British Columbia government said it retained B.C. and California counsel to explore legal action against OpenAI over alleged failure to notify law enforcement before the February 10 Tumbler Ridge shooting. The province says flagged ChatGPT threats were not reported before the attack, which killed eight people and wounded 27; OpenAI has previously said its threshold for law-enforcement referral was not met. For AI safety and policy teams, the case is a concrete escalation point: moderation review, repeat-offender detection, emergency referral thresholds, and cross-border legal exposure are moving from trust-and-safety playbooks into government litigation strategy.
OpenAI made ChatGPT for PowerPoint generally available for ChatGPT Business workspaces on July 6, 2026, and its office-agent tools are moving toward token-based credit pricing. The feature runs inside Microsoft PowerPoint so teams can create, revise and review editable decks while workspace admins control app access. OpenAI's help docs also say Workspace Agent and Excel/Sheets tasks now use credits based on input, cached-input and output tokens, with PowerPoint remaining free for Business and Enterprise customers through August 6, 2026. For practitioners, the main shift is governed AI assistance moving into the delivery surface for board decks, QBRs and analysis summaries.
ABC reported from Australia on July 8 that a Department of Employment and Workplace Relations report found AI is not yet causing broad disruption to Australia's labour market, based on occupational changes since ChatGPT launched in November 2022. The report found labour-market conditions remain strong by historical standards while also noting early weakness in some occupations predicted to be highly exposed, including telemarketers. Employment and Workplace Relations Minister Amanda Rishworth said AI could still reshape jobs, and ABC reports the government plans regular monitoring to detect future changes. For practitioners, the useful signal is methodological: broad layoff narratives should be tested against occupation-level trends, vacancy data, task exposure, and repeated public baselines rather than one-off anecdotes.
