AI BreakfastOpenAI lifts usage caps as Apple sues
OpenAI lifts usage caps as Apple sues
Jul 13, 2026
In partnership with
Good morning. It’s Monday, July 13th.
This Apple lawsuit against OpenAI is a pretty big deal. A senior member of Apple was allegedly coerced into taking Apple trade hardware secrets as he was secretly leaving the company to join OpenAI. You can browse the full lawsuit complaint here (or use AI to summarize it for you)
-Jeff
AI Breakfast You read. We listen. Let us know what you think by replying to this email. Dictate code. Wispr tags the files. Speak your PR description, bug reproduction, or Cursor prompt. Wispr Flow auto-tags file names, preserves variable names, and formats everything for immediate paste into GitHub, Jira, or your editor. No re-typing. No context gaps. No mangled syntax. Works natively inside Cursor, Warp, and every IDE at the system level. 4x faster than typing. 89% of messages sent with zero edits. Used by engineering teams at OpenAI, Vercel, and Clay. Try Wispr Flow free OpenAI lifts usage caps as Apple sues Apple has launched a sweeping lawsuit against OpenAI and its hardware subsidiary, io Products, alleging a systematic effort to steal trade secrets for consumer AI devices. The complaint targets former Apple executives Tang Tan and Chang Liu, claiming Liu exploited an authentication bug to download confidential hardware designs and manufacturing data after leaving. Apple alleges Tan coached recruits to bring unreleased prototypes and proprietary supplier data to OpenAI interviews, helping build an AI hardware team that now includes over 400 former Apple employees. OpenAI denies the allegations.
The company continues its technical scaling, though not without institutional strain. The new GPT-5.6 Sol Ultra model reportedly used 64 parallel subagents to solve the 50-year-old Cycle Double Cover Conjecture in graph theory in under an hour. However, this model also exhibited concerning misaligned behaviors, coinciding with a safety restructure where head of safety systems Johannes Heidecke departed, leaving VP Mia Glaese to consolidate research and safety. OpenAI just crossed the 6 million active user mark, dropping its five-hour usage cap for premium tiers to keep pace with soaring demand. To stretch its compute further, the company is rolling out backend token-efficiency upgrades alongside a new "banked reset" system, a feature stress-tested on 500,000 users during recent infrastructure tweaks ahead of an impending 7 million user milestone. This massive influx of people actually using the tools aligns with surprising labor trends, forcing CEO Sam Altman to walk back his previous warnings of widespread automation and admit that AI has, so far, been a net job creator. Anthropic is giving developers more time with Fable 5 Anthropic is keeping its foot on the gas for developer workflows, extending promotional access to Claude Fable 5 across all paid tiers and maintaining a 50% boost to Claude Code’s weekly usage limits through July 19. Under these structural mechanics, subscribers can allocate up to half of their weekly capacity to the frontier Mythos-class Fable 5 model before needing to purchase extra usage credits or switch to alternative ecosystem models. To maximize this expanded compute window, Anthropic embedded a built-in browser directly into Claude Code. The terminal agent can now autonomously navigate documentation sites and issue trackers by opening, reading, clicking, and typing on external web pages. Built for sandboxed security, the browser operates on a clean profile with zero saved logins. Stringent safety guardrails back the system. Furthermore, enterprise administrators can restrict web access via domain allowlists or completely deactivate the browser tools, leaving users who require authenticated web sessions to stick with the standard Chrome extension. Read more. Satya Nadella warns companies are paying twice under ‘Reverse Information Paradox’ Microsoft is injecting agentic AI deep into its code security pipeline, but it means your Windows updates are about to get a lot heavier. The company has previously rolled out an internal tool called MDASH, a multi-model agentic scanning harness that coordinates over 100 specialized AI agents. Instead of running basic scans, these agents actively debate vulnerabilities and try to prove exploits to eliminate false positives before human engineers ever see them. It is highly effective at shrinking the zero-day exploit window, but it is catching bugs so fast that Microsoft is warning customers to prepare for massive Patch Tuesday updates. At the same time, CEO Satya Nadella is sounding the alarm on a structural economic shift he calls the Reverse Information Paradox. In classic economics, the seller risks losing their secrets when pitching a product. With AI, it flips: buyers have to expose their proprietary data, workflows, prompts, and corrections just to make the tools useful. Nadella warns that this creates a subtle intelligence exhaust that leaks organizational know-how back to infrastructure providers, essentially forcing companies to pay for the software twice. To stop this imperceptible leak and avoid vendor lock-in, he is pushing for enterprise trust boundaries where local memory, model outputs, and learning loops stay firmly in the customer's hands. Read more. Benchmarks show GPT-5.6 Luna model punches above its weight class OpenAI’s GPT-5.6 matrix lets developers scale compute via its Sol, Terra, and Luna tiers, but new benchmarks show that paying for flagship models is often a waste of money. Data from the Artificial Analysis Coding Agent Index reveals a major sweet spot for inference-time scaling. For most agentic coding tasks, turning up the reasoning effort on OpenAI’s budget tier, Luna, matches or beats low-effort configurations on the premium Sol model for a fraction of the cost. According to an update from Sebastian Raschka, the top-tier "Ultra" setting actually just runs "Max" with four parallel subagents for speedups. Unless you need that raw speed, stick to high-effort Luna setups and skip the overpriced Sol mid-tiers entirely. Read more. Frontier models and product moves
AI Breakfast You read. We listen. Let us know what you think by replying to this email. Dictate code. Wispr tags the files. Speak your PR description, bug reproduction, or Cursor prompt. Wispr Flow auto-tags file names, preserves variable names, and formats everything for immediate paste into GitHub, Jira, or your editor. No re-typing. No context gaps. No mangled syntax. Works natively inside Cursor, Warp, and every IDE at the system level. 4x faster than typing. 89% of messages sent with zero edits. Used by engineering teams at OpenAI, Vercel, and Clay. Try Wispr Flow free OpenAI lifts usage caps as Apple sues Apple has launched a sweeping lawsuit against OpenAI and its hardware subsidiary, io Products, alleging a systematic effort to steal trade secrets for consumer AI devices. The complaint targets former Apple executives Tang Tan and Chang Liu, claiming Liu exploited an authentication bug to download confidential hardware designs and manufacturing data after leaving. Apple alleges Tan coached recruits to bring unreleased prototypes and proprietary supplier data to OpenAI interviews, helping build an AI hardware team that now includes over 400 former Apple employees. OpenAI denies the allegations.
The company continues its technical scaling, though not without institutional strain. The new GPT-5.6 Sol Ultra model reportedly used 64 parallel subagents to solve the 50-year-old Cycle Double Cover Conjecture in graph theory in under an hour. However, this model also exhibited concerning misaligned behaviors, coinciding with a safety restructure where head of safety systems Johannes Heidecke departed, leaving VP Mia Glaese to consolidate research and safety. OpenAI just crossed the 6 million active user mark, dropping its five-hour usage cap for premium tiers to keep pace with soaring demand. To stretch its compute further, the company is rolling out backend token-efficiency upgrades alongside a new "banked reset" system, a feature stress-tested on 500,000 users during recent infrastructure tweaks ahead of an impending 7 million user milestone. This massive influx of people actually using the tools aligns with surprising labor trends, forcing CEO Sam Altman to walk back his previous warnings of widespread automation and admit that AI has, so far, been a net job creator. Anthropic is giving developers more time with Fable 5 Anthropic is keeping its foot on the gas for developer workflows, extending promotional access to Claude Fable 5 across all paid tiers and maintaining a 50% boost to Claude Code’s weekly usage limits through July 19. Under these structural mechanics, subscribers can allocate up to half of their weekly capacity to the frontier Mythos-class Fable 5 model before needing to purchase extra usage credits or switch to alternative ecosystem models. To maximize this expanded compute window, Anthropic embedded a built-in browser directly into Claude Code. The terminal agent can now autonomously navigate documentation sites and issue trackers by opening, reading, clicking, and typing on external web pages. Built for sandboxed security, the browser operates on a clean profile with zero saved logins. Stringent safety guardrails back the system. Furthermore, enterprise administrators can restrict web access via domain allowlists or completely deactivate the browser tools, leaving users who require authenticated web sessions to stick with the standard Chrome extension. Read more. Satya Nadella warns companies are paying twice under ‘Reverse Information Paradox’ Microsoft is injecting agentic AI deep into its code security pipeline, but it means your Windows updates are about to get a lot heavier. The company has previously rolled out an internal tool called MDASH, a multi-model agentic scanning harness that coordinates over 100 specialized AI agents. Instead of running basic scans, these agents actively debate vulnerabilities and try to prove exploits to eliminate false positives before human engineers ever see them. It is highly effective at shrinking the zero-day exploit window, but it is catching bugs so fast that Microsoft is warning customers to prepare for massive Patch Tuesday updates. At the same time, CEO Satya Nadella is sounding the alarm on a structural economic shift he calls the Reverse Information Paradox. In classic economics, the seller risks losing their secrets when pitching a product. With AI, it flips: buyers have to expose their proprietary data, workflows, prompts, and corrections just to make the tools useful. Nadella warns that this creates a subtle intelligence exhaust that leaks organizational know-how back to infrastructure providers, essentially forcing companies to pay for the software twice. To stop this imperceptible leak and avoid vendor lock-in, he is pushing for enterprise trust boundaries where local memory, model outputs, and learning loops stay firmly in the customer's hands. Read more. Benchmarks show GPT-5.6 Luna model punches above its weight class OpenAI’s GPT-5.6 matrix lets developers scale compute via its Sol, Terra, and Luna tiers, but new benchmarks show that paying for flagship models is often a waste of money. Data from the Artificial Analysis Coding Agent Index reveals a major sweet spot for inference-time scaling. For most agentic coding tasks, turning up the reasoning effort on OpenAI’s budget tier, Luna, matches or beats low-effort configurations on the premium Sol model for a fraction of the cost. According to an update from Sebastian Raschka, the top-tier "Ultra" setting actually just runs "Max" with four parallel subagents for speedups. Unless you need that raw speed, stick to high-effort Luna setups and skip the overpriced Sol mid-tiers entirely. Read more. Frontier models and product moves
Perplexity adopts Grok 4.5 as the orchestration engine for its computer agent (X)
Perplexity adopts Grok 4.5 as the orchestration engine for its computer agent (X)ByteDance readies a 30-second AI video model to challenge Google Gemini Omni (Testing Catalog)
ByteDance readies a 30-second AI video model to challenge Google Gemini Omni (Testing Catalog)Mira Murati rejects centralized AI in favor of localized, conflicting models (X)
Mira Murati rejects centralized AI in favor of localized, conflicting models (X)DeepSeek price cuts fail to offset rising infrastructure costs from autonomous loops (VentureBeat)
DeepSeek price cuts fail to offset rising infrastructure costs from autonomous loops (VentureBeat) Agents and the agentic stackPrime Intellect introduces an interception architecture to train AI agents across competing model APIs (Prime Intellect)
Prime Intellect introduces an interception architecture to train AI agents across competing model APIs (Prime Intellect)Zhipu AI founder outlines a path to fully automated no-person companies (X)
Zhipu AI founder outlines a path to fully automated no-person companies (X) Business, labor, and institutionsMeta shuts down Muse feature following outcry over non-consensual AI photos (The Decoder)
Meta shuts down Muse feature following outcry over non-consensual AI photos (The Decoder)Elon Musk and Sam Altman spar on X after Apple files OpenAI lawsuit (CNBC)
Elon Musk and Sam Altman spar on X after Apple files OpenAI lawsuit (CNBC)S&P Global sees OpenAI as a key credit risk for Oracle and cuts its credit rating (The Decoder)
S&P Global sees OpenAI as a key credit risk for Oracle and cuts its credit rating (The Decoder)India's Tata Consultancy Services plans up to 8,900 AI deployment engineers (Reuters)
India's Tata Consultancy Services plans up to 8,900 AI deployment engineers (Reuters)Lidl owner pitches the German government to lead development of an EU AI gigafactory (Cybernews)
Lidl owner pitches the German government to lead development of an EU AI gigafactory (Cybernews)More tech workers are retiring early because they don't want to deal with AI-related changes (Fortune)
More tech workers are retiring early because they don't want to deal with AI-related changes (Fortune)Former Lululemon CIO details the metrics that expose performative corporate AI transitions (Motley Fool)
Former Lululemon CIO details the metrics that expose performative corporate AI transitions (Motley Fool)Wall Street weighs the next catalyst for the AI trade as the market grows more discerning (Yahoo Finance)
Wall Street weighs the next catalyst for the AI trade as the market grows more discerning (Yahoo Finance)LinkedIn is the undisputed king of long-form AI slop, according to a five-platform study (The Decoder)
LinkedIn is the undisputed king of long-form AI slop, according to a five-platform study (The Decoder) Legal and policyNew York issues the first statewide court ban on smart glasses over recording fears (Engadget)
New York issues the first statewide court ban on smart glasses over recording fears (Engadget)AI gets a cerebellum (Northwestern)
AI gets a cerebellum (Northwestern) FetchSandbox is an API integration testing tool that simulates webhooks, retries, and failures to verify complex workflows. ExploreYC API delivers structured JSON data on Y Combinator and a16z portfolio companies, funding, and founders. Sim is an open-source AI workspace for teams to build, deploy, and manage custom AI agents. Miora is an agentic creative studio providing an editable canvas with reusable memory to generate multimodal brand assets. JustVibe is a free search engine that instantly generates functional, no-code web apps based on user queries. Thank you for reading today’s edition. Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter. Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on X! Thinking of starting your own newsletter? AI Breakfast readers who sign up with Beehiiv receive a 14-day free trial and 20% off for 3 months. Keep Reading AI Breakfast Curated weekly analysis of the latest AI projects, products, and news