Apple Sues OpenAI for Trade Secret Theft New York Times / Bloomberg / TechCrunch Apple filed suit in the Northern District of California accusing OpenAI of a systematic, multi-level campaign to steal hardware trade secrets — from job candidates being told to bring physical Apple prototypes to interviews, to a former engineer who exploited a post-departure bug to keep accessing Apple’s cloud storage. The lawsuit targets OpenAI hardware chief Tang Tan, a former Apple VP, as the alleged orchestrator and seeks an injunction to block OpenAI from possessing any Apple IP. The filing marks a stunning reversal from the 2024 ChatGPT-on-iPhone partnership.


Apple Sues OpenAI, Accusing It of Stealing Company SecretsNew York Times Apple’s complaint alleges misconduct “at every level” of OpenAI’s hardware division: job interviewers requesting unreleased device components for “show and tell,” a former Apple engineer laughing in Slack messages about accessing Apple cloud storage after leaving the company, and outreach to Apple’s manufacturing partners to learn proprietary metal-finishing techniques. The lawsuit follows OpenAI’s $6.4 billion acquisition of Jony Ive’s IO Products and the subsequent freeze in the Apple-OpenAI partnership.

The 6 Wildest Claims in Apple’s Lawsuit Against OpenAIThe Verge The Verge unpacks the most eye-catching allegations: former Apple VP Tang Tan directing candidates to smuggle hardware, a bug-exploit that gave a departed engineer ongoing cloud access, and OpenAI approaching Apple’s supply-chain partners to reverse-engineer finishing processes.

The Wildest Allegations in Apple’s Trade Secrets Lawsuit Against OpenAITechCrunch (companion roundup of the complaint’s standout claims)

Sam Altman’s Space Data Center Trash TalkTechCrunch Elon Musk accused Altman of being a scammer; Altman fired back publicly about Musk “selling space datacenter” schemes to public-market investors. Most infrastructure analysts already agree that space-based data centers are unviable near-term, giving Altman’s jab factual grounding if not diplomatic tone.

Your Azure DevOps Pipelines Have a 2027 DeadlineSysadmin Central — Azure DevOps retires its legacy workload identity federation issuer on July 1, 2027; older service connections must migrate to Microsoft Entra or break.


🤖 Frontier Models

Gemini 3.5 Pro Targets July 17 After Full Architectural RebuildTechTimes Google DeepMind scrapped the Gemini 2.5 Pro architecture entirely to compete with GPT-5.6 and Fable 5, rebuilding for improvements in mathematical reasoning, SVG generation, and image quality. The rebuilt model is expected to ship a 2M-token context window and a “Deep Think” reasoning layer, though no specs are officially confirmed ahead of the Thursday target.

“The Wave Has Arrived”: Zhipu Co-Founder Tang Jie’s Letter to StaffGeoPolitechs In a rare, unusually candid public letter, Zhipu’s co-founder announces the company is returning fully to foundation-model research with a two-year AGI roadmap covering long-horizon tasks, autonomous agents, full self-training, and what he calls “extreme safety governance.” A useful window into Chinese AI lab strategic thinking.

Google, Microsoft, Salesforce Back New AI Agent Interoperability Standard to Counter MCPAIToolsRecap — A consortium including Google, Microsoft, Salesforce, Snowflake, and ServiceNow is backing a shared standard for connecting AI agents to business software, positioning it against Anthropic’s widely adopted Model Context Protocol.

Pepsi Challenge for LLMsThread — LLMs have matured to the point where existing benchmarks no longer capture meaningful differences; the field needs new evaluation frameworks.


🏢 Enterprise & Product

Anthropic Starts Localizing Claude Pricing for IndiaTechCrunch Claude users in India are now seeing Indian rupee-denominated subscription plans, a signal that Anthropic is treating India as a serious growth market rather than an incidental one. India is Anthropic’s largest user base outside the US.

AI-Generated Code Creates 78% More IncidentsNew Relic (2026 State of AI Coding Report) New Relic’s report finds that while most teams rate AI code quality as higher than human-written code at review time, shipped AI-generated code leads to 78% more production incidents. 62% of engineering leaders say their teams ship AI code without line-by-line verification, and 96% say observability is now imperative.

Code Review Is Theater NowOctopus Deploy AI-driven PR volume now outpaces human reviewer capacity, making traditional code review an increasingly ineffective quality gate. The argument: code review still has value for mentorship and design discussion, but safety guarantees have shifted downstream to automated pipelines and policy checks.

Working at the Frontier: How Hebbia Builds AI for Financial DiligenceClaude Blog Hebbia’s case study on building zero-miss financial diligence systems with Claude, covering how they navigate the reliability requirements that distinguish high-stakes enterprise AI from consumer use cases.

Waze Adds Gemini AI-Powered FeaturesThe Verge — Google integrates Gemini into Waze for personalized trip planning, updated conversational reporting, and reduced chattiness, competing more directly with Apple Maps.

The Reverse Information ParadoxSatya Nadella / X — Nadella argues AI creates a structural problem where buyers must reveal proprietary knowledge to make models useful, enriching the vendor asymmetrically; companies need ways to use AI without that data loss.


🔧 Developer Tools & Infrastructure

pgrust: A Rust Rewrite of PostgreSQL Reaches Postgres 18.3 CompatibilityGitHub pgrust passes all 46,000+ Postgres regression queries and boots from existing Postgres data directories. The developers claim 50% faster transaction throughput and ~300x faster analytical workloads versus standard Postgres. The project uses a thread-per-connection model instead of Postgres’s traditional process-per-connection architecture — a significant departure worth watching.

Highlights from Git 2.55GitHub Blog Git 2.55 ships incremental multi-pack index repacking to reduce repository maintenance overhead, plus new history fixup capabilities for moving staged changes into earlier commits.

Databricks Feature Views — Public PreviewDatabricks — Define ML features once; Databricks automatically handles training and real-time serving pipelines with 200ms streaming latency from Kafka, integrated with Unity Catalog.

Cloudflare Smart Tiered Cache for Public Cloud RegionsCloudflare — Free update that lets customers specify cloud region hints so Cloudflare avoids transcontinental cache routing when origins sit behind anycast IPs from AWS, GCP, Azure, or Oracle Cloud.

Open-Inspect: Open-Source Background Coding Agent SystemGitHub — Open-source recreation of Ramp’s Inspect, supporting automated refactoring, bug fixes, and documentation across multiple repos with Slack and CLI integrations.

How to Protect Your MCP Server with FastlyFastly — Fastly’s prototype for MCP server security combines edge networking, bot management, and WAF rules; a 10-week trial met enterprise latency and throughput targets.

Jenkins Smart RetryJenkins — Policy-based pipeline retries that classify failures first, recovering from transient issues like lost agents or SCM interruptions without blindly retrying deterministic failures.


🤖 Hardware & Robotics

Apple’s M6, M7, and M8 Chips Show How AI Is Reshaping the CompanyBloomberg Mark Gurman reports that AI is no longer a feature Apple’s chips must support — it is now the primary driver of chip design and product timing decisions. Notably, AI hardware developed for the since-cancelled Apple Car project now powers Macs and Apple’s AI servers, reframing the car’s cancellation as an expensive but consequential R&D investment.

1X’s New Robotic Hands Try to Solve the Part That Actually MattersThe Next Web 1X’s Neo robot gets 25-joint tendon-driven hands with pressure-and-shear sensing skin that can detect a glass starting to slip. The hands can bend past human range to handle irregular shapes — targeting the core unsolved problem of home robotics: the enormous variability of household objects and placements.

BrainCo Bets on Wearable Brain Tech While Neuralink Drills Into SkullsThe Next Web — Hangzhou-based BrainCo’s EEG headbands read electrical signals through the scalp without surgery, positioning non-invasive BCI as the consumer path even if it captures fewer channels than implants.

China Recovered Its First Reusable RocketArs Technica — Long March 10B’s first stage was caught by a sea-based net catch on its maiden flight, a novel recovery method distinct from SpaceX’s approach, and the opening of China’s reusable launch capability.


Generated by claude-sonnet-4-6 · 2026-07-13T10:00:00Z