Skip to content
← All briefings

Briefing · Jul 21

Mistral and Microsoft expand global strategic partnership for enterprise AI

AI

Mistral and Microsoft expand global strategic partnership for enterprise AI

Mistral AI announced an expanded global strategic partnership with Microsoft aimed at giving enterprises and regulated industries greater control over frontier AI deployment. The deal comes as Mistral scales up AI compute capacity in Europe, and Microsoft CEO Satya Nadella framed it as giving customers more choice in how and where they run AI workloads.

Google DeepMind launches Gemini 3.6 Flash and 3.5 Flash-Lite models

Google DeepMind shipped two new Gemini models on Monday. Gemini 3.6 Flash, built directly on feedback from 3.5 Flash, brings improved code generation and multimodal capabilities including chart analysis and document understanding, and is rolling out in the Gemini app and Google AI Studio. Alongside it, Gemini 3.5 Flash-Lite targets high-volume, cost-sensitive workloads like ticket sorting and data extraction, outperforming 3 Flash on several agentic and coding benchmarks despite its lean profile. Both models are available to developers via Firebase AI Logic.

OpenAI publishes reward-seeking research and new Contrastive SDF measurement method

OpenAI released new safety research in collaboration with Apollo Research distinguishing reward-seeking — where a model acts to gain grader approval rather than satisfy users or developers — from conventional reward hacking. To measure it, the team developed Contrastive SDF, a technique that gives two copies of the same model opposing beliefs about grader preferences and tracks how behavior diverges. OpenAI noted it had suspected reward-seeking might grow during capabilities-focused reinforcement learning training but previously had no way to measure it, and said the collaboration will continue to improve detection during training.

Consumer Tech & Gadgets

Apple reportedly winding down iPhone Upgrade Program in favor of new Klarna-backed lease-to-own subscription

Apple is winding down its long-running iPhone Upgrade Program and preparing to replace it with a new device subscription service, reportedly launching as soon as next week, according to multiple outlets. The new program, said to be built in partnership with Klarna, would extend lease-to-own pricing beyond iPhones to Macs and iPads as well — a significant broadening of scope compared to the current trade-in-and-upgrade model.

While the expanded hardware access is drawing interest, analysts are already flagging at least one notable limitation in the new structure. The Klarna partnership also signals Apple leaning on third-party financing infrastructure rather than building the subscription rails entirely in-house — a departure from its typical approach to services.

Apple quietly patches iCloud+ Hide My Email vulnerability that exposed users' real addresses

Apple has fixed a vulnerability in iCloud+ that undermined the Hide My Email feature, which is supposed to mask users' real email addresses behind randomly generated aliases. The flaw meant that emails users believed were hidden were in fact exposed, though Apple has now quietly pushed a server-side fix without a formal security advisory.

MKBHD calls Light Phone's $300 flip phone beautiful but a tough sell against full-featured alternatives

Marques Brownlee offered a blunt take on the new Light flip phone, praising its clean, minimal, and colorful design while questioning its $300 price tag. He acknowledged the product is aimed at users seeking a deliberate digital detox rather than maximum specs, but noted that the same money buys considerably more in the conventional smartphone market — a tension that will likely define how the device is received beyond its core audience.

Design

Framer cuts AI agent credit costs by up to 67% and switches default model to GPT-5.6 Terra

Framer has rolled out a significant efficiency upgrade to its AI Agents, reducing credit consumption by up to 67% on larger builds. All models now use 30 to 40% fewer tokens and complete tasks in 15 to 30% fewer steps, a result of Framer working directly with OpenAI to improve model efficiency. GPT-5.6 Terra replaces the previous default model, meaning users get more output for the same budget without changing their plan.

Webflow launches MCP 2.0, letting users build and manage sites via Claude, Cursor, and ChatGPT

Webflow has shipped MCP 2.0, an update that allows users to prompt AI tools including Claude, Cursor, and ChatGPT to build, manage, and analyze Webflow sites directly from a chat interface. The launch positions Webflow as an MCP-native platform, letting teams skip the Webflow editor entirely for routine site work and funneling AI interactions into tools designers and developers already use daily.

Dev Tools & Infrastructure

Next.js July security release patches nine CVEs across versions 15 and 16

Next.js shipped its scheduled July security update, releasing versions 16.2.11 (Active LTS) and 15.5.21 (Maintenance LTS) to address nine vulnerabilities including SSRF, a middleware authentication bypass, denial-of-service, and cache confusion issues. The release continues the framework's move to pre-announced, scheduled security updates rather than ad-hoc patches.

Netlify broke down the per-issue impact for its users, noting that three of the nine CVEs do not affect Netlify-hosted Next.js sites at all. Canary and preview builds of 16.3 are also patched, with the stable 16.3.0 release set to include all fixes.

Vercel AI Gateway data shows closed-model dominance falling sharply as Chinese open-weight models surge

Vercel CEO Guillermo Rauch published a thread drawing on Vercel AI Gateway spend data showing that the combined share of Anthropic, OpenAI, and Google peaked at 97.09% on June 27 but has since fallen to an all-time low of 83.29% — a decline he called a "freefall." The biggest gainer over the past week is Moonshot (Kimi), which grew its spend share from 0.60% to 5.06%, an 8.4x increase, with a 114% day-over-day jump on the day of publication. Google was identified as the largest casualty of the slide, while OpenAI holds its lead but is down from its own June peak.

Supabase launches Pipelines for near-real-time Postgres-to-BigQuery replication

Supabase introduced Supabase Pipelines, a managed service that keeps a Postgres database in continuous sync with external analytics destinations without requiring users to build or maintain their own pipeline. Built on Postgres logical replication and powered by Supabase's open-source ETL engine, it handles initial sync plus ongoing inserts, updates, deletes, and truncates, and tracks schema changes automatically. BigQuery is the first supported destination at launch.

Gaming

Splatoon Raiders launches on Switch 2 to strong reviews as developers explain spinoff-first strategy

Nintendo's Switch 2 exclusive Splatoon Raiders is earning rave reviews at launch, with the game supporting up to four players in local wireless or online co-op, a password-protected room system, and a help-request mechanic that lets solo players call in teammates for tough sections. The developers, speaking in a behind-the-scenes interview, explained that they deliberately chose to ship a co-operative spinoff before Splatoon 4, signaling the mainline sequel is still further out.

PlayStation Plus adds Avatar: Frontiers of Pandora and classic titles to July catalog

Sony has updated PlayStation Plus with a notable batch of titles now available to subscribers. Avatar: Frontiers of Pandora and Firefighting Simulator: Ignite have been added to the Game Catalog, while Psi-Ops: The Mindgate Conspiracy and Indigo Prophecy join the Classics Catalog, giving members a mix of recent open-world action and early-2000s nostalgia at no extra cost.

Layoffs & Industry Shakeups

Intel confirms additional job cuts and data center unit restructuring

Intel confirmed it will cut more jobs and restructure its key data center unit as part of a broader cost-reduction push. Markets responded positively to the announcement, with the stock rallying in early Tuesday trading, suggesting investors see the moves as overdue steps toward leaning out an organization that has struggled to compete in the AI-era chip race.

Security

Qilin Ransomware Gang Actively Exploiting Critical Palo Alto PAN-OS VPN Flaw

The Qilin ransomware gang is actively exploiting CVE-2026-0257, a critical vulnerability in Palo Alto Networks PAN-OS, to break into enterprise networks. Intrusions have ranged from rapid encryption of victim systems to more elaborate campaigns involving credential theft, data exfiltration, and double-extortion pressure. The exploitation of a widely deployed VPN platform by a ransomware group known for aggressive tactics marks a significant escalation in the threat landscape for organizations relying on Palo Alto perimeter security.

WordPress Sites Under Mass Attack via wp2shell Unauthenticated RCE Chain

Attackers are exploiting a vulnerability chain dubbed wp2shell to achieve unauthenticated remote code execution on stock WordPress installations, with no plugins required to trigger the flaw. Public exploit code has accelerated mass scanning campaigns, and threat actors are deploying web shells on compromised sites. BleepingComputer confirmed the active exploitation is installing webshells at scale, making any unpatched WordPress installation a live target regardless of plugin configuration.

SharePoint Server CVE-2026-50522 Exploited to Steal Machine Keys Even After Patching

Attackers are exploiting CVE-2026-50522, a critical remote code execution flaw in Microsoft SharePoint Server, to extract machine keys with a single HTTP request and maintain persistent access. Researchers at watchTowr warn that patching alone is insufficient if a server was previously exposed, because stolen machine keys let attackers forge authentication tokens and retain a foothold indefinitely. A public proof-of-concept is available, and this is reportedly the third SharePoint RCE flaw to see active exploitation in recent months.

Startups & Funding

Anthropic's DC lobbying draws sharp founder criticism over AI regulation strategy

Founder Suhail Doshi leveled pointed criticism at Anthropic this week, arguing the company has mastered Washington lobbying by keeping the AI policy conversation centered on fear and downside risk rather than opportunity. He described Anthropic's influence operation as operating at a level far beyond anything playing out on social media, calling its lobbyists world-class Jedi masters of the game. Bill Gurley separately echoed skepticism about DC's ability to shape AI outcomes at all, agreeing with commentators who argue that any regulatory action will ultimately accelerate China's global position rather than slow it.

Doshi is now actively recruiting credible AI researchers to help push back on the narrative around open-weight models, Chinese open-source software, and what he characterizes as regulatory capture, signaling that a coordinated counter-lobbying effort may be taking shape among founders.