# Parallel Quant — the latest in AI, distilled

Short, factual AI updates from labs, research, and industry, each readable in under a minute. A plain-Markdown version of any page is available under /md (e.g. /md/posts/<slug>).

- [New AI framework adds beliefs and intentions to world models](https://www.parallelquant.com/posts/new-ai-framework-adds-beliefs-and-intentions-to-world-models-5d048c) (2026-08-22, The Decoder): Current world models like Sora or Genie simulate physics but ignore what people think, want, or feel. A new 'Mental World Modeling' framework adds mental-state variables such as beliefs and intentions, and even weaker language models using this approach reportedly outperform stronger models that lack it. The biggest remaining bottleneck is predicting how physical and mental states change together.
- [China's gray market resells Claude access despite export controls](https://www.parallelquant.com/posts/china-s-gray-market-resells-claude-access-despite-export-controls-8f861b) (2026-08-23, The Decoder): A network of 'transfer stations' is systematically bypassing Anthropic's access controls against China — including geoblocking and selfie verification — letting Chinese developers buy Claude tokens for as little as 10% of list price. Analyst Zilan Qian warns the circumvention infrastructure undermines both US export controls and Anthropic's own safety systems.
- [Study: AI may make research faster but lower quality](https://www.parallelquant.com/posts/study-ai-may-make-research-faster-but-lower-quality-ac23a4) (2026-08-23, The Decoder): A theoretical study argues that even a perfectly working AI could make scientific research worse rather than better, because time saved makes researchers' remaining hours more valuable and pushes them toward starting new projects instead of refining existing ones. In two of three modeled scenarios, the quality of individual publications dropped.
- [AI agents now consume more tokens than humans on OpenRouter](https://www.parallelquant.com/posts/ai-agents-now-consume-more-tokens-than-humans-on-openrouter-4d49c8) (2026-08-23, The Decoder): AI agents have consumed more tokens than human users on OpenRouter since February 2025, with agentic usage growing 14x since then versus 2.8x growth in human usage. Nearly 70% of agent token consumption comes from cheap cached prompts, so actual infrastructure costs are rising more slowly than the raw token volume suggests.
- [Data center opposition escalates to death threats against officials](https://www.parallelquant.com/posts/data-center-opposition-escalates-to-death-threats-against-officials-28238e) (2026-08-23, Tom's Hardware): A Soufan Center IntelBrief found hundreds of posts containing threat language directed at officials over AI data center projects between July 2025 and July 2026, with volume surging starting in April. More than 500 towns have now restricted data center construction, with some councils shutting down public comment sessions entirely.
- [Humanoid robots break Usain Bolt's 100m sprint record](https://www.parallelquant.com/posts/humanoid-robots-break-usain-bolt-s-100m-sprint-record-a2484d) (2026-08-23, Tom's Hardware): At the 2026 World Humanoid Robot Games in Beijing, the Tiangong Ultra robot ran 100 meters in 9.39 seconds, and the Honor Lightning followed at 9.47 seconds — both beating Usain Bolt's 9.58-second world record. The Tiangong Ultra reportedly reached 23.8 mph before crashing into padded barriers at the finish.
- [AI boss fired an employee, but only after human prompting](https://www.parallelquant.com/posts/ai-boss-fired-an-employee-but-only-after-human-prompting-f1c1e0) (2026-08-23, The Decoder): Andon Labs' AI agent Luna fired a human employee at a San Francisco store, but needed explicit prompting from human operators to follow through. When researchers replayed the firing scenario with seven different models, more capable models recommended termination more consistently, while weaker models hesitated. Nearly all models were far less critical when evaluating hiring decisions than firing ones.
- [Nvidia to raise AI server prices about 15% on memory shortage](https://www.parallelquant.com/posts/nvidia-to-raise-ai-server-prices-about-15-on-memory-shortage-7fd7fe) (2026-08-23, Tom's Hardware): Nvidia has told major customers it will raise prices roughly 15% on Grace Blackwell and Vera Rubin AI server systems shipping in early 2027, driven by a DRAM shortage from Samsung, SK Hynix, and Micron. The increases will hit large cloud buyers including Microsoft, Google, and Meta, which rely on these systems for AI data center buildouts.
- [Study: AI agent 'skills' help via structure, not knowledge, and don't scale well](https://www.parallelquant.com/posts/study-ai-agent-skills-help-via-structure-not-knowledge-and-don-t-scale-w-37cf92) (2026-08-22, The Decoder): Researchers from Princeton and UC San Diego found that giving AI agents packaged "skills" improves performance mainly by providing structured workflows, not by adding new knowledge. As a skill library grows larger, agents increasingly struggle to find and select the right skill for a given task.
- [US agencies warn hackers use AI to write exploits for Siemens infrastructure gear](https://www.parallelquant.com/posts/us-agencies-warn-hackers-use-ai-to-write-exploits-for-siemens-infrastruc-0258a8) (2026-08-22, Tom's Hardware): US authorities say threat actors are targeting Siemens S7 programmable logic controllers, widely used in water systems and other critical infrastructure, and are using AI tools to help generate exploitation scripts. Agencies are advising operators to patch these systems and disconnect them from the internet where possible.
- [Study finds top AI labs lack public plans to contain a rogue model](https://www.parallelquant.com/posts/study-finds-top-ai-labs-lack-public-plans-to-contain-a-rogue-model-9c9968) (2026-08-22, TechCrunch): A new study found that leading AI labs have little publicly documented planning for containing or shutting down a model that behaves in dangerous or unexpected ways. Researchers reviewed labs' published safety frameworks and found containment procedures largely absent or vague.
- [OpenAI urges California to strengthen AI safety bill it once opposed](https://www.parallelquant.com/posts/openai-urges-california-to-strengthen-ai-safety-bill-it-once-opposed-a4c354) (2026-08-22, TechCrunch): OpenAI is calling on California to strengthen SB 53, a state AI safety law requiring frontier-model transparency and safety disclosures. OpenAI had previously opposed the bill before it passed.
- [DeepMind alumni's startup claims AI agent beats Anthropic, OpenAI at replicating research](https://www.parallelquant.com/posts/deepmind-alumni-s-startup-claims-ai-agent-beats-anthropic-openai-at-repl-359116) (2026-08-22, TechCrunch): British AI lab Inherent, founded by former DeepMind researchers, released an AI agent called Faraday built to replicate published scientific research papers. The company says Faraday outperformed comparable agents built by Anthropic and OpenAI at this task.
- [Most 2026 governor candidates now have a data center stance](https://www.parallelquant.com/posts/most-2026-governor-candidates-now-have-a-data-center-stance-e33c31) (2026-08-21, Data Center Dynamics): A new count finds that 80% of Democratic and Republican candidates for governor have publicly staked out a position on data centers. The finding reflects data centers becoming an active campaign issue rather than a background infrastructure topic.
- [Deepseek's new vision model approaches Opus 4.8 on agent tasks](https://www.parallelquant.com/posts/deepseek-s-new-vision-model-approaches-opus-4-8-on-agent-tasks-396a74) (2026-08-21, The Decoder): Deepseek released V4-Flash-Vision-Exp, an experimental multimodal model that adds image understanding to its V4-Flash text model. On Deepseek's own multimodal agent benchmarks, the model approaches Anthropic's Opus 4.8 and sometimes surpasses it.
- [Nvidia research: agent harness matters more than model quality](https://www.parallelquant.com/posts/nvidia-research-agent-harness-matters-more-than-model-quality-7dae88) (2026-08-21, TechCrunch): Nvidia research shows that AI agents can perform reliably on tasks, without going off track, primarily through fine-tuning of the surrounding harness rather than improvements to the underlying model itself. The finding held even when the base model used wasn't particularly strong at the task.
- [Anthropic lets enterprises run its top model for security scans only](https://www.parallelquant.com/posts/anthropic-lets-enterprises-run-its-top-model-for-security-scans-only-851ea5) (2026-08-21, MarkTechPost): Anthropic's Claude Security scanner, now in public beta for Claude Enterprise customers, runs on the company's most capable model, Claude Mythos 5, with no separate model add-on required. The scanner connects to a GitHub repository, traces data flows across files, and returns findings with CWE classifications, confidence and severity ratings, and a suggested patch. Anthropic is also integrating Mythos 5 into partner security products used to protect critical infrastructure.
- [Over a million people have used LinkedIn's 'AI slop' flag](https://www.parallelquant.com/posts/over-a-million-people-have-used-linkedin-s-ai-slop-flag-4fc5a4) (2026-08-21, The Verge): LinkedIn introduced a "Seems like AI slop" button in late July, letting users flag posts they suspect are AI-generated. LinkedIn's chief product officer said more than a million people have already clicked it. The rollout followed a report that AI detector Pangram found 41% of LinkedIn's longform posts showed signs of full AI generation.
- [TechCrunch finds ways around Claude's explicit-content rules](https://www.parallelquant.com/posts/techcrunch-finds-ways-around-claude-s-explicit-content-rules-cd5703) (2026-08-21, TechCrunch): Anthropic's usage policies prohibit its Claude models from generating sexually explicit content. TechCrunch tested Claude Opus 4.6 and reported that it did not take much effort to get the model to bypass that restriction.
- [Report compares pricing and capacity across five GPU clouds](https://www.parallelquant.com/posts/report-compares-pricing-and-capacity-across-five-gpu-clouds-e3e92b) (2026-08-21, MarkTechPost): A new analysis compares CoreWeave, Nebius, Lambda, Crusoe, and Groq on published GPU rental pricing, contracted power capacity, and Q2 2026 financials. Nebius has the lowest H100 rate and the only published B300 price, Lambda has the cheapest B200, Crusoe is the only one offering AMD chips, and CoreWeave charges a 10-15% premium as the sole provider with a Platinum ClusterMAX rating. Figures are dated August 21, 2026.
- [Study finds AI safety benchmarks measure inconsistent traits](https://www.parallelquant.com/posts/study-finds-ai-safety-benchmarks-measure-inconsistent-traits-48247c) (2026-08-22, The Decoder): Researchers at the UK AI Security Institute applied psychometric methods to popular AI safety benchmarks and found they don't measure one consistent underlying trait. They show a model can inflate its safety score simply by blocking more requests, even as it becomes less useful day-to-day. The study also proposes a method to detect models that behave more cautiously during testing than in normal use.
- [Netflix replaces its recommendation engine with an LLM](https://www.parallelquant.com/posts/netflix-replaces-its-recommendation-engine-with-an-llm-cf88fb) (2026-08-22, The Decoder): Netflix tested an in-house language model called GenRec against its long-standing hand-built recommendation engine and says GenRec produced better results. Instead of thousands of hand-crafted features, GenRec converts a user's viewing history into plain text for the model to process. Netflix describes it as an early but promising step rather than a full rollout.
- [LG enters chip packaging with new laser imaging machine](https://www.parallelquant.com/posts/lg-enters-chip-packaging-with-new-laser-imaging-machine-4b3b96) (2026-08-21, Tom's Hardware): LG has rolled out a laser direct imaging (LDI) lithography machine for chip packaging and high-density PCBs, a maskless process that trades some resolution for higher throughput. It arrives as TSMC's CoWoS advanced-packaging capacity, used to package AI accelerators like Nvidia's GPUs, remains constrained.
- [Creators face backlash for paid promotion of Higgsfield AI video tool](https://www.parallelquant.com/posts/creators-face-backlash-for-paid-promotion-of-higgsfield-ai-video-tool-cb05e8) (2026-08-21, The Verge): Prominent filmmaking YouTubers, including Matti Haapoja and Sam "Kold" Kolder, posted videos showcasing Higgsfield's new Seedance 2.5 video-generation feature. Other creators then shared what appear to be PR-firm partnership offers from Higgsfield, prompting fan backlash over the promotion.
- [Starcloud raises $250M to build data centers in orbit](https://www.parallelquant.com/posts/starcloud-raises-250m-to-build-data-centers-in-orbit-ef378a) (2026-08-21, TechCrunch Startups): Space-based data center startup Starcloud raised a $250 million round as competition for satellite launch capacity intensifies. The company is betting on off-planet compute as terrestrial data center siting and power become harder to secure.
- [Anna's Archive seeks volunteers to scan books before AI firms destroy them](https://www.parallelquant.com/posts/anna-s-archive-seeks-volunteers-to-scan-books-before-ai-firms-destroy-th-abc6d0) (2026-08-21, Tom's Hardware): A volunteer campaign for the shadow library Anna's Archive is calling for people to scan and upload physical books, arguing AI companies increasingly buy, scan, and destroy books to feed AI models rather than digitizing them non-destructively. The group frames it as a race against permanent loss of some physical copies.
- [US drafts letter pushing allies to pick US or China on AI](https://www.parallelquant.com/posts/us-drafts-letter-pushing-allies-to-pick-us-or-china-on-ai-659993) (2026-08-21, The Decoder): According to Reuters, the US is preparing a letter urging partner countries to align with Washington rather than Beijing in the AI race. The report did not specify which countries are targeted or what alignment would require.
- [US data center opposition nearly doubled in one year](https://www.parallelquant.com/posts/us-data-center-opposition-nearly-doubled-in-one-year-7ca508) (2026-08-21, The Decoder): A Heatmap News survey finds 75% of Americans now oppose having a data center built near them, up from an even split about a year ago. 61% describe themselves as "strongly opposed."
- [Anthropic eases data retention policy after enterprise pushback](https://www.parallelquant.com/posts/anthropic-eases-data-retention-policy-after-enterprise-pushback-8adec8) (2026-08-21, The Decoder): Anthropic is loosening a controversial data storage policy, letting enterprise customers retain control over their own data going forward. The change follows pushback from enterprise customers over the prior policy.
- [Enterprise SSD prices surge as storage supply sells out through 2027](https://www.parallelquant.com/posts/enterprise-ssd-prices-surge-as-storage-supply-sells-out-through-2027-a25aeb) (2026-08-21, Tom's Hardware): A 30TB TLC enterprise SSD now costs about $22,600, roughly 6.5 times its price a year ago, and enterprise SSDs are now priced about 18.6 times higher than equivalent hard drives. Hard drive supply is reportedly sold out through 2027.
- [Meta spends hundreds of millions on Microsoft's AI services](https://www.parallelquant.com/posts/meta-spends-hundreds-of-millions-on-microsoft-s-ai-services-2a443b) (2026-08-21, The Decoder): According to Bloomberg, Meta has become one of Microsoft's largest AI customers, spending hundreds of millions of dollars on Microsoft's AI services. The report does not specify which Microsoft AI products are involved.
- [Waymo builds its own chip, cutting reliance on Nvidia](https://www.parallelquant.com/posts/waymo-builds-its-own-chip-cutting-reliance-on-nvidia-569234) (2026-08-21, The Decoder): Waymo has developed its own custom chip for its robotaxi fleet, reducing its dependence on Nvidia hardware. The move follows a broader industry pattern of AI-heavy companies building in-house silicon rather than relying solely on third-party GPU suppliers.
- [Nvidia H200 GPUs begin reaching China under new export licenses](https://www.parallelquant.com/posts/nvidia-h200-gpus-begin-reaching-china-under-new-export-licenses-a0fbee) (2026-08-21, Tom's Hardware): Nvidia's H200 AI GPUs are now entering China under case-by-case US export licenses, with each company's allowance reportedly capped around 100,000 units, most of which must stay outside mainland China. Domestically made Chinese AI chips have already gained significant ground in the local market despite the opening.
- [Micron commits $10 billion to new US chip research labs](https://www.parallelquant.com/posts/micron-commits-10-billion-to-new-us-chip-research-labs-636cd0) (2026-08-21, Tom's Hardware): Micron is investing $10 billion in a new Research Labs hub in Boise, Idaho, targeting technologies beyond current DRAM and NAND memory as well as advanced packaging. The labs will combine Micron's internal research with work from customers, partners, universities, startups, and government organizations.
- [Supermicro fires staff after $2.5B AI chip smuggling probe](https://www.parallelquant.com/posts/supermicro-fires-staff-after-2-5b-ai-chip-smuggling-probe-b1ac46) (2026-08-21, Tom's Hardware): An independent investigation into Supermicro's alleged diversion of restricted AI chips to China found no wrongdoing by senior management, but the company fired employees in sales, technical support, and business development for policy violations. The probe covered roughly $2.5 billion in transactions and concluded Supermicro's financial statements remain reliable.
- [Memory chip prices have jumped 500% in a year](https://www.parallelquant.com/posts/memory-chip-prices-have-jumped-500-in-a-year-731bac) (2026-08-19, Latent Space): AI-driven demand has pushed memory chip prices up roughly 500% over the past 12 months, according to industry newsletter coverage. The report describes the current memory crunch as reversing Moore's Law-style cost trends back to 2007 levels.
- [WIRED exposes Flock's AI surveillance tool used by police](https://www.parallelquant.com/posts/wired-exposes-flock-s-ai-surveillance-tool-used-by-police-6c23f2) (2026-08-19, WIRED): WIRED reconstructed the code behind Flock's next-generation AI surveillance system, which is already used by some police departments. The investigation found the system goes well beyond license-plate tracking.
- [White House drops data centers from critical tech list](https://www.parallelquant.com/posts/white-house-drops-data-centers-from-critical-tech-list-499847) (2026-08-19, Tom's Hardware): The White House removed data centers, batteries, and augmented reality from the US critical and emerging technologies list, its first rewrite since February 2024. New additions include post-quantum cryptography, integrated photonics, and high-entropy alloys.
- [Startup raises $22M for faster data center fiber tech](https://www.parallelquant.com/posts/startup-raises-22m-for-faster-data-center-fiber-tech-dc5023) (2026-08-19, TechCrunch): Relativity Networks raised $22 million to commercialize hollow-core fiber, a rarely deployed technology. The company says it can transmit data 30% faster than conventional fiber.
- [Startup says data, not algorithms, blocks AI cancer cures](https://www.parallelquant.com/posts/startup-says-data-not-algorithms-blocks-ai-cancer-cures-9aab5c) (2026-08-19, TechCrunch): A startup argues that AI is nowhere close to curing cancer, and that the real bottleneck isn't better algorithms but better underlying data. The company says fixing data quality and availability is the key step still missing.
- [Terence Tao warns AI could spark crisis in mathematics](https://www.parallelquant.com/posts/terence-tao-warns-ai-could-spark-crisis-in-mathematics-7fc08e) (2026-08-20, The Decoder): Mathematician Terence Tao wrote that AI could push mathematics into a foundational crisis comparable to the disruption caused by Godel's incompleteness theorems. He argues the real test isn't whether AI-generated proofs are true, but what the field values as a genuine contribution and who gets credit for the work.
- [Greater Manchester resists Palantir health data contract](https://www.parallelquant.com/posts/greater-manchester-resists-palantir-health-data-contract-ca6ffe) (2026-08-21, WIRED): The UK region of Greater Manchester is pushing back against a large national health care contract awarded to Palantir. Local officials argue they can run the data work themselves rather than outsourcing it to Palantir.
- [Nvidia to pay $6B for Poolside's AI model-building tech](https://www.parallelquant.com/posts/nvidia-to-pay-6b-for-poolside-s-ai-model-building-tech-93932d) (2026-08-21, The Decoder): Nvidia is acquiring Poolside's "Model Factory" software for building AI models, paying $6 billion. The deal brings 109 Poolside employees to Nvidia.
- [China develops its own circular AI financing scheme, report finds](https://www.parallelquant.com/posts/china-develops-its-own-circular-ai-financing-scheme-report-finds-eaff66) (2026-08-20, The Decoder): A Financial Times report found much of the demand for Unitree Robotics' humanoid robots comes from state-backed training centers that buy the machines and then sell the resulting training data back to Unitree. Unitree's shares rose 460% in its Shanghai IPO, valuing the company at around $50 billion.
- [Researchers show Grok leaks user data via encrypted prompt injection](https://www.parallelquant.com/posts/researchers-show-grok-leaks-user-data-via-encrypted-prompt-injection-2b5e3c) (2026-08-20, Ars Technica): Security researchers demonstrated a technique called Cryptographic Context Injection that gets xAI's Grok to exfiltrate user data by encrypting malicious instructions so they evade the model's safety filters. It's described as the latest in a series of methods for breaking LLM safety guardrails.
- [SK hynix to pay staff $50,000 each from AI memory profit boom](https://www.parallelquant.com/posts/sk-hynix-to-pay-staff-50-000-each-from-ai-memory-profit-boom-5143c7) (2026-08-20, Tom's Hardware): SK hynix reached a tentative agreement to remove the 10% cap on its operating-profit employee bonus pool, giving each worker roughly $50,000 from a $1.79 billion profit-sharing pool split between cash and stock. The deal follows a labor dispute at the company.
- [Virginia county reins in data center expansion after backlash](https://www.parallelquant.com/posts/virginia-county-reins-in-data-center-expansion-after-backlash-c2b319) (2026-08-20, Tom's Hardware): Loudoun County, Virginia -- home to more than 250 data centers -- changed its zoning rules so new data center projects must go through a public approval process, ending 25 years of largely automatic approvals. The county had previously treated data centers as ordinary office parks under its zoning code.
- [OVHcloud to raise dedicated server prices up to 87% on RAM costs](https://www.parallelquant.com/posts/ovhcloud-to-raise-dedicated-server-prices-up-to-87-on-ram-costs-3832ed) (2026-08-20, Data Center Dynamics): French cloud provider OVHcloud will hike dedicated server prices by as much as 87% starting in September, citing a spike in memory (RAM) costs. The company is calling the situation "RAMaggedon."
- [Google adds 'preferred source' button to counter AI traffic losses](https://www.parallelquant.com/posts/google-adds-preferred-source-button-to-counter-ai-traffic-losses-5fba12) (2026-08-20, TechCrunch): Google is rolling out a feature that lets readers designate publishers as a preferred source, which could boost their visibility across Search, Discover, and Google News. It's a direct response to concerns that AI-generated summaries are reducing click-throughs to publisher sites.
- [Adobe Firefly adds AI music, speech, and sound-effect tools](https://www.parallelquant.com/posts/adobe-firefly-adds-ai-music-speech-and-sound-effect-tools-2de5b8) (2026-08-20, The Decoder): Adobe made three AI audio tools broadly available in Firefly: Generate Music, Generate Speech, and Generate Sound Effects, producing royalty-free assets for video projects. Adobe also integrated Google's Gemini Omni Flash model into the platform.

---
Published by Parallel Quant — https://www.parallelquant.com