AI Product Updates Daily — June 6, 2026

OpenAI: Dreaming V3 heads to free tier, GPT-5.5-Cyber expands to EU

OpenAI's most substantive memory overhaul since ChatGPT launched went live on June 4 for Plus and Pro users in the US, and is now rolling out to free-tier and international users in the weeks ahead. Dreaming V3 replaces the old saved-memories list with a background synthesis process: after each conversation ends, the system reads across years of chat history, extracts what matters — preferences, constraints, active projects, time-sensitive context — and updates your profile without a "remember this" prompt. If you mentioned flying to Singapore in July, the system will eventually update that entry to past-tense rather than surfacing it forever. OpenAI says the new architecture is approximately five times more compute-efficient than its predecessor, which is what makes the free-tier expansion financially viable; Plus and Pro users also get double the memory storage.1

openai.comhttps://openai.com/index/chatgpt-memory-dreaming/External link

Loading content card…

The catch: privacy researchers flagging the system note that a February 2026 arXiv study found 96% of ChatGPT memories in a sample of 2,050 entries were created unilaterally by the system, without any user prompt. EU AI Act transparency rules take effect in August 2026 and could require explicit disclosure of what gets stored.2

On the enterprise front, OpenAI is extending GPT-5.5-Cyber access to vetted EU teams — cybersecurity teams, businesses, government authorities, and EU institutions including the EU AI Office. The move is a direct competitive shot at Anthropic: UK banks were excluded from Project Glasswing's latest expansion and are being offered GPT-5.5-Cyber instead.3 Japan's three megabanks are also receiving access through the same verified-defender program.4

Anthropic: Glasswing at 150 orgs, IPO in SEC review, outage resolved

Anthropic's Project Glasswing — its program giving vetted organizations access to Claude Mythos Preview for cybersecurity applications — expanded its partner network on June 2 to cover power grids, water systems, healthcare networks, communications infrastructure, and hardware manufacturers. The new additions bring the total to over 150 organizations across 15 countries, with Anthropic estimating the combined codebases support systems affecting more than 100 million people.5 Separately, Anthropic connected Claude to 28 security and compliance platforms through its Claude Compliance API in late May, with integrations spanning CrowdStrike, Palo Alto Networks, Okta, and Zscaler.6

The Claude Opus 4.8 that launched May 28 at $5/$25 per million tokens is now available across all major platforms: Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry (see also the Google Cloud section below).7 Opus 4.8 is notable for explicitly signaling model uncertainty — a deliberate design choice — and for extended agentic coding sessions with dependency tracking.

Developer circles are still watching for Claude Sonnet 4.8, which has not been officially announced. The evidence rests on a source map accidentally shipped in the @anthropic-ai/claude-code npm package v2.1.88 on March 31, 2026, containing the string sonnet-4-8. Since Opus 4.7 subsequently shipped as the same leak predicted, a mid-June Sonnet release is widely anticipated — but treat it as unconfirmed.8

Anthropic's confidential S-1 filing with the SEC (submitted June 1) is now in formal review. Revenue run-rate hit approximately $47 billion in May — roughly 5x the prior year. The $65B Series H at a $965B valuation was completed days before the filing. No price, ticker, or timing has been set; this is the SEC review phase. One notable line item from the filing: Anthropic will pay SpaceX $1.25 billion per month through May 2029 for compute.8

A brief Claude outage on Friday, June 5 (elevated error rates across several Opus models) was resolved within two hours.9

Google Cloud: Claude Opus 4.8 added to Gemini Enterprise Agent Platform

Google Cloud confirmed that Claude Opus 4.8 is now available on the Gemini Enterprise Agent Platform — giving enterprise customers access to the model through Vertex AI for multi-stage agentic workflows and extended coding sessions.10 Google had previously flagged June 17, 2026 as the date when Search Console AI opt-out settings go live — publishers who block their content from Google AI Overviews will see it take effect then (the Gemini app is explicitly excluded from that opt-out).11

Apple: WWDC26 keynote on June 8 — rebuilt Siri, iOS 27, and AI Extensions

Apple's developer conference begins June 8 at 10 a.m. PT with what Bloomberg and AppleInsider describe as the company's "AI redemption" moment. The expected highlights:12 13

x.comhttps://x.com/AppleExternal link

Loading content card…

Rebuilt Siri: Two years late, the overhauled Siri is expected to launch in beta form, occupying the top of the iPhone display via an expanded Dynamic Island. It will support multi-step commands and context awareness, with Google Gemini as the underlying on-device intelligence layer.
Siri Extensions: iOS 27 reportedly includes a dedicated App Store section for third-party AI models, letting users choose which AI service powers Siri — a modular architecture that would let OpenAI, Claude, or others run as Siri back-ends.
Apple Intelligence: Across iOS 27, iPadOS 27, and macOS 27, including Photos expansions (extend, enhance, recompose), a rebuilt Camera UI with a Visual Intelligence button, auto tab grouping in Safari, and Wallet improvements for splitting receipts.
iPhone Fold support: iOS 27 is expected to include Split View functionality adapted for the foldable display.

The broader software update cycle is positioned as a stability and polish release — less overhaul than 2025's iOS 26, more Snow Leopard than Leopard.

Microsoft: canceling 10,000+ internal Claude Code licenses

Microsoft is shutting down its internal Claude Code licenses across its Experiences and Devices division by June 30, redirecting thousands of engineers toward GitHub Copilot. The shift follows an internal assessment that token costs for Claude Code had spiraled beyond budget. Microsoft invested $18 billion into OpenAI and has been building its Copilot infrastructure to handle the same workloads.8 14

The cost pressure is visible across the ecosystem. Supabase raised $500M at a $10.5B valuation in a Series F led by Singapore's GIC, with more than 60% of new databases on the platform now started by AI coding agents (Claude Code is the largest single source).15 The valuation doubled in eight months.

Regulation: Great American AI Act, Trump EO, and Colorado deadline at 25 days

Three regulatory actions are shaping the enterprise AI conversation this week.

Great American AI Act (discussion draft, June 4): Representatives Jay Obernolte (R-CA) and Lori Trahan (D-MA) released a 269-page bipartisan draft proposing a three-year federal preemption of state AI laws related to frontier model development. Companies with over $500M in annual revenue would be required to publish Frontier AI Frameworks, report critical safety incidents, and support a $100M/year federal AI standards center. Labor unions (AFL-CIO, AFT) issued a joint rejection; tech industry groups praised it. The draft is open for comment — no formal introduction yet.16

Trump AI & Cybersecurity EO (June 2/5): President Trump signed "Promoting Advanced Artificial Intelligence Innovation and Security," directing voluntary collaboration between AI developers and the federal government on cyber threat detection. CISA is developing a new platform to help agencies use AI for defensive purposes. The EO is described by analysts as light-touch, prioritizing acceleration over binding requirements.17

Colorado AI Act (June 30): The Colorado Consumer Protections for Artificial Intelligence Act — the first comprehensive state AI law in the US — takes effect in 25 days. It applies anti-discrimination requirements to high-risk AI systems in employment, healthcare, financial services, education, housing, and legal services. If the Great American AI Act passes, it would preempt Colorado's law for three years. Companies serving Colorado residents need compliance frameworks whether or not the federal bill moves.8

NVIDIA RTX Spark and Cognizant Physical AI platform

NVIDIA RTX Spark, the Arm-based superchip announced June 1 at Computex, continues to draw enterprise attention. Consumer laptops using the chip are expected in autumn 2026; Adobe is rebuilding Photoshop and Premiere Pro around the architecture. Jensen Huang has positioned this as NVIDIA's move to own AI workloads at the edge — agents running locally, without cloud latency.18

Cognizant launched a Sovereign Physical AI Platform-as-a-Service on June 5, built on its Intelligence Spine architecture. The platform connects industrial sensors, IoT devices, factory automation, and energy infrastructure into a unified system across eight core verticals. The announcement came alongside an expanded CrowdStrike strategic alliance for AI lifecycle security.19

One more: Kling AI turns two

Kuaishou's Kling AI is celebrating its second anniversary today, June 6. Over two years the team released 26 model iterations and expanded globally. No new model has been announced at the time of publication, but the anniversary is historically paired with capability drops.

Loading content card…