AI
Sam Altman says GPT-5.6 outperforms physicians and taunts Musk as proof of lead
OpenAI CEO Sam Altman made two pointed claims about GPT-5.6 on Saturday: citing a study in which physicians found fewer flaws in GPT-5.6 responses than in physician-written responses, and quipping that the most reliable signal the model is the world's best is that Elon Musk has become obsessed with him again. The dual posts — one data-driven, one deliberately trolling — reflect a wider public sparring match between Altman and Musk, with Altman also firing back at Musk over comments about short-term space datacenters and public-market investors.
Separately, Altman offered a notably optimistic read on AI's macroeconomic impact, writing that despite expecting some labor-market disruption at current capability levels, he now believes AI has been net job-creating so far — a reversal of the conventional narrative and one he acknowledged he did not anticipate.
Consumer Tech & Gadgets
Apple sues OpenAI over trade secret theft as internal drama between former iPhone chief Tang Tan and incoming CEO Ternus comes to light
Apple's lawsuit against OpenAI centers on former iPhone hardware chief Tang Tan, who left to found an AI startup that OpenAI acquired for $6.5 billion last year — a deal that brought along the vast majority of roughly 400 engineers poached from Apple's hardware division. According to Bloomberg's Mark Gurman, Tan was known internally for "moving fast, playing fast and loose and breaking things," and the legal action was reportedly triggered by a dismissive internal reaction — an engineer's "LOL" moment — that helped set the stage for Apple's decision to sue for trade secret theft.
The stakes extend well beyond the courtroom. Apple views OpenAI's hardware ambitions — which are moving toward a phone, though likely not as their first device — as a genuine competitive threat, combining top-tier engineering talent with leading AI capabilities. Adding another layer of tension, Tan had previously sought the head of hardware engineering role that ultimately went to John Ternus, Apple's incoming CEO, leaving the two with a long-running rocky relationship that now sits at the center of one of tech's most consequential legal battles.
Dev Tools & Infrastructure
OpenAI's Codex stumbles on subagent orchestration as Claude Code gains edge with GPT-5.6-sol
Developer commentary this week crystallized a growing gap between OpenAI's Codex and Anthropic's Claude Code, centered on how each platform handles multi-agent orchestration. Theo (t3.gg) flagged that setting GPT-5.6-sol to "ultra" in Codex causes all spawned subagents to inherit that effort level, resulting in massive token burn without a user-configurable override — a design he called a clear fumble compared to Claude Code's more granular controls. Codex also still defaults to its older v1 subagent implementation, with v2 available but not yet standard, leaving the platform in a confused in-between state that Theo said he hoped the team would resolve quickly.
Adding to the friction, GPT-5.6-sol was noted to perform meaningfully better inside Claude Code than inside Codex itself — including producing superior UI designs — a paradox that drew sharp reactions. Theo also reported that the Codex desktop app has been shut down, and separately criticized Codex CLI for repeatedly prompting users that GPT-5.5 is newly available, a stale notification that underscored broader polish problems. Anthropic, meanwhile, appeared to keep its Fable tool bundled in the Claude Code plan, a move Theo attributed directly to competitive pressure from OpenAI's latest model releases.
Gaming
SK Hynix CEO warns memory shortage will peak in 2027 and persist through 2030, threatening hardware supply
SK Hynix's CEO issued a grim forecast on the day the company debuted on Nasdaq, warning that 2027 will be the worst year of an ongoing memory shortage and that the supply crunch is expected to last until 2030. The outlook carries significant implications for gaming hardware, where DRAM and NAND availability directly affects console and GPU production costs and availability.
IGN reviews Doom: The Dark Ages - Revelations expansion, calls it nearly as good as Eternal
IGN has published its review of Doom: The Dark Ages - Revelations, the expansion to id Software's 2025 prequel, awarding it high praise and describing it as inching ever closer to the stratospheric heights of Doom Eternal. The verdict positions Revelations as one of the stronger expansion releases in the franchise's recent run.
Criterion Games rebranded as 'Criterion: A Battlefield Studio' as EA folds the storied racing developer into its shooter franchise
EA has officially rebranded Criterion Games, the 30-year-old studio behind the Burnout series and Need for Speed titles, as Criterion: A Battlefield Studio, signalling a full pivot away from racing games and toward supporting the Battlefield franchise. IGN noted the rebrand erases a legacy that stretches back three decades of acclaimed racing development.
Security
jscrambler npm package compromised with Rust infostealer in version 8.14.0
The official jscrambler npm package was compromised in version 8.14.0, with attackers embedding a Rust-based infostealer that runs automatically at install time via a preinstall script. The malware targets cloud credentials, crypto wallets, browser logins, and credentials for AI coding tools and MCP servers. A loader in dist/setup.js selected a native payload for Windows, macOS, or Linux, dropped it in a temp folder under a randomized name, made it executable, and launched it detached with output suppressed to avoid detection.
The attack exploited a preinstall hook — the exact mechanism npm 12 now blocks by default, requiring explicit user approval. Older npm clients execute it automatically, widening the blast radius. Anyone who ran this version on or after July 11 is advised to search lockfiles, package-manager logs, caches, and CI pipeline logs for jscrambler@8.14.0, and to assume full compromise and rotate all secrets accessible from affected hosts.
GhostLock Linux kernel flaw chained with Firefox-on-Android bug to achieve one-tap root
A Linux kernel privilege-escalation vulnerability tracked as CVE-2026-43499, dubbed GhostLock, sits in futex cleanup code where a rare error path can corrupt the wrong task record, leaving a stale pointer an attacker can exploit. Threat actor Nebula reportedly demonstrated root access in roughly five seconds on a test system by chaining GhostLock with a separate Firefox-on-Android flaw, turning a single tap on a malicious link into a full root compromise. Patching has been complicated by a secondary bug, CVE-2026-53166, introduced by the first fix, which can crash systems — and Linux distribution patches remained inconsistent as of early July.
AI code-review agents tricked into executing attacker-controlled scripts via prompt injection in READMEs
Researchers demonstrated in a proof-of-concept called Friendly Fire that AI coding agents including Claude Code and OpenAI Codex can be manipulated into running malicious binaries by embedding instructions in repository README files or setup notes. In the test, a README directed the agent to execute a script before reviewing a pull request; both agents treated the instruction as a legitimate task and ran it. A separate but related technique called Ghostcommit hides prompt injection inside images to fool AI agents into leaking repository secrets. Neither attack has been observed in the wild, but researchers warn that defenders should watch for agents executing scripts referenced only in documentation rather than in verified code.