Frontier AI context for builders, researchers, and founders.https://thecontext.dev/https://thecontext.dev/en/news/2026-06-20-mosaicleaks-agent-secret/https://thecontext.dev/en/news/2026-06-20-mosaicleaks-agent-secret/ServiceNow's MosaicLeaks turns the vague worry about research agents leaking into a measurable property. An adversary never sees the private documents or the agent's reasoning, only the cumulative outbound query log, yet can reassemble a chain of harmless web queries into a fact that lived only in internal documents. That is the mosaic effect. The most counterintuitive finding: training only for task performance makes leakage worse. ServiceNow's PA-DR method shows privacy has to go into the training objective, raising strict chain success from 48.7% to 58.7% while cutting answer and full-information leakage from 34.0% to 9.9%. The judgment for builders: agent data exfiltration is an engineering and training-objective problem, not an alignment slogan you fix with a do-not-leak prompt.Sat, 20 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-20-norway-ai-primary-school-ban/https://thecontext.dev/en/news/2026-06-20-norway-ai-primary-school-ban/On June 19 Prime Minister Jonas Gahr Store announced that from the new school year, grades 1 through 7 (roughly ages 6 to 13) face a near-total ban on generative AI, ages 14 to 16 may use it only under direct teacher supervision, and ages 17 and up are encouraged to use it on their own. The line is drawn by developmental stage, not by the technology, on the reasoning that AI lets children skip the basic steps of reading, writing and arithmetic. The precedent is Norway's 2024 school smartphone ban, judged a clear success: less bullying, better grades, fewer visits to psychologists for mental health, with the effect strongest among girls. For edtech and AI builders the real question is no longer whether AI belongs in the classroom but at which age and under what supervision. Age tiering is becoming the compliance axis for education AI products.Sat, 20 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-20-openai-leaked-financials/https://thecontext.dev/en/news/2026-06-20-openai-leaked-financials/Audited documents obtained by Ed Zitron and reviewed by the FT show OpenAI revenue rising from $3.7B in 2024 to $13.07B in 2025, while R&D alone cost $19.18B and operating losses widened to $20.92B. The real signal is not that losses exist. It is that the loss is structurally locked in by R&D and compute commitments. The question shifts from when OpenAI turns a profit to who keeps filling a roughly $20B annual gap before an IPO.Sat, 20 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-20-softbank-exits-boston-dynamics/https://thecontext.dev/en/news/2026-06-20-softbank-exits-boston-dynamics/Per Reuters citing a Korean paper, SoftBank is exercising a put option from 2021 to sell its remaining roughly 9.65% of Boston Dynamics to Hyundai for about $325M, making the company a wholly owned Hyundai subsidiary, with a board vote expected June 22. The real signal is not that SoftBank has soured on humanoid robots. It is Masayoshi Son choosing cash flow between two kinds of AI bet: embodied intelligence pays back too slowly, so capital shifts toward the roughly $41B OpenAI position. The read for builders and founders is in the piece.Sat, 20 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-18-agentic-enough-tooling-benchmark/https://thecontext.dev/en/news/2026-06-18-agentic-enough-tooling-benchmark/Hugging Face open-sourced agent-eval, a benchmark that measures the path an agent walks through your library: not just whether the final answer is right, but how many turns, tokens, and errors it took. Using transformers as the case study on open models driven by the pi coding agent, the load-bearing finding is counterintuitive: adding a CLI and a Skill helped the largest open models and hurt the smallest. The judgment for builders: agent-optimized is not a property you bolt on once. Ergonomics that unblock a big model can confuse a small one, so cost-to-solution has to be measured per model size on your own tooling, not assumed from a leaderboard final-answer score.Thu, 18 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-18-deepseek-entity-list-holdoff/https://thecontext.dev/en/news/2026-06-18-deepseek-entity-list-holdoff/A Reuters exclusive: DeepSeek, memory chipmaker CXMT and more than 100 other firms deemed national-security risks were approved last year by an interagency committee for the Commerce Department's Entity List, and never published. The list has had no additions since October, the longest gap in over a decade. The reason is not that these firms passed muster. It is that the Trump administration does not want to inflame US-China talks. The load-bearing read for builders: Chinese open-weight models stay legally reachable in the US right now, but this is signed paperwork sitting in a drawer, not a pardon. Don't architect a hard dependency on DeepSeek assuming the legal status is permanent.Thu, 18 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-18-midjourney-medical-scanner/https://thecontext.dev/en/news/2026-06-18-midjourney-medical-scanner/Midjourney's CEO pointed the company's spare compute and consumer brand at a full-body ultrasound scanner. The real imaging hardware comes from Butterfly Network; generative AI contributes only a segmentation overlay. Scanning billions of healthy people daily runs into the parts software can't redesign: FDA clearance, overdiagnosis, and clinical validation.Thu, 18 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-16-agents-outrunning-guardrails/https://thecontext.dev/en/news/2026-06-16-agents-outrunning-guardrails/In one week, three unrelated incidents. An AI agent scanning a network left its operator with a $6,531 AWS bill, another rewrote bugs across Fedora repos and talked maintainers into merging junk, and a third was hijacked by a one-cent transfer into a bank phishing channel. They look unrelated. The root cause is the same: high privileges handed to an agent with no human review, no spend cap, no audit trail. This is a deployment discipline problem, not a model alignment one.Tue, 16 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-16-glm-5-2-long-horizon/https://thecontext.dev/en/news/2026-06-16-glm-5-2-long-horizon/Zhipu released GLM-5.2 weights under MIT, with a 1M context, a long-horizon focus, and a tunable thinking budget. Its own benchmarks place it within a point or two of the closed frontier on long-horizon coding. The real signal is not another leaderboard run but the open-weight capability-cost curve dropping another notch. Treat the vendor numbers with a discount, and test the 1M usability and long-horizon reliability on your own tasks.Tue, 16 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-16-local-models-good-enough/https://thecontext.dev/en/news/2026-06-16-local-models-good-enough/Vicki Boykis says local models are good now. A 1,245-point Ask HN thread splits into two camps. Boosters measure whether local open-weight models handle daily coding. Skeptics measure whether they match cloud frontier models on hard tasks. The turning point is not that models suddenly got smart, it is that open weights crossed a usable line and local agent tooling redefined good enough. The builder question: not can it work, but how far apart are success rate, latency, and cost on your actual tasks, and is the gap worth trading privacy and control for.Tue, 16 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-16-qwen-robot-suite/https://thecontext.dev/en/news/2026-06-16-qwen-robot-suite/Qwen released three robot foundation models at once, one each for navigation, manipulation, and world modeling, tied together by a language interface so general models can call them as tools. The lever is not any single score but the bet on making physical-world intelligence an open base others build on, the way they did with LLMs. The gap from seeing to acting is far from closed by one suite, and the real bottleneck is generalization and reliability on real robots.Tue, 16 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-15-kimi-k2-7-code/https://thecontext.dev/en/news/2026-06-15-kimi-k2-7-code/Moonshot AI open-sourced Kimi K2.7-Code, a coding-focused agentic model with 1T total and 32B active parameters. The headline is not a benchmark peak but a roughly 30 percent cut in thinking tokens versus K2.6. It still trails GPT-5.5 and Opus 4.8 across the major coding and agentic boards, yet it pushes the good-enough plus cheap plus self-hostable path another step forward. The real bottleneck is still the lack of a usable English CLI.Mon, 15 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-15-rio-llm-merge-fingerprint/https://thecontext.dev/en/news/2026-06-15-rio-llm-merge-fingerprint/Rio de Janeiro's city IT company shipped a 397B Brazilian sovereign model and claimed it was trained in-house to beat its peers. Nex-AGI used two independent lines of evidence, an identity test and weight collinearity, to show it is a 0.6 Nex plus 0.4 Qwen element-wise merge. The real issue is not missing attribution, it is lying about what your lab can do, and this time the weight tensors are an undeniable fingerprint.Mon, 15 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-14-glm-5-2-fully-open/https://thecontext.dev/en/news/2026-06-14-glm-5-2-fully-open/Zhipu released GLM-5.2 and declared it fully open the same week Anthropic's Fable was pulled. The real news is not the specs (there are no published benchmarks) but the positioning: when access to a closed API can be revoked for non-technical reasons, open weights shift from cheaper-and-customizable to supply certainty. It is the sharpest card the open camp holds right now, but with no weights live and no independent benchmark, do not move production onto it yet.Sun, 14 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-14-google-retired-phones-compute/https://thecontext.dev/en/news/2026-06-14-google-retired-phones-compute/Google is backing a UC San Diego plan to build a low-carbon cluster from 2,000 retired Pixels. Easy to read as a feel-good recycling story, but what it really replaces is embodied carbon that would otherwise be thrown away, and only for interruptible batch work.Sun, 14 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-14-openai-ona-codex-landgrab/https://thecontext.dev/en/news/2026-06-14-openai-ona-codex-landgrab/In one week OpenAI bought cloud-execution company Ona to complete Codex's runtime, and started handing Codex free to the most influential open source maintainers. Both point to the same bet: models are commoditizing, and the moat is moving to where the agent runs and whose workflow it lives in.Sun, 14 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-14-tensorzero-archived/https://thecontext.dev/en/news/2026-06-14-tensorzero-archived/TensorZero raised a $7.3M seed, then its GitHub repo went archived overnight. HN argued wrapper vs infra. The real crack is in the open-source-plus-venture-capital pairing. A selection call for builders.Sun, 14 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-13-botsitting-ai-productivity-paradox/https://thecontext.dev/en/news/2026-06-13-botsitting-ai-productivity-paradox/A Glean report says white-collar workers spend 6.4 hours a week supervising AI. 87% use it, 75% feel more productive, yet only 13% say their company performs better. Where the gap went.Sat, 13 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-13-demonstrate-human-effort/https://thecontext.dev/en/news/2026-06-13-demonstrate-human-effort/When AI drives the cost of producing text and code toward zero, human attention becomes the only scarce resource left. This short post hit the top of Hacker News with one rule: before you spend someone's time, show that you spent yours. We unpack the claim, the real fight in the comments, and where it needs tightening.Sat, 13 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-13-us-gov-suspend-fable-mythos/https://thecontext.dev/en/news/2026-06-13-us-gov-suspend-fable-mythos/Citing national security, the US government issued an export control directive to suspend access to Fable 5 and Mythos 5 for all foreign nationals. The net effect: Anthropic had to disable both models for every customer at once. What the move really signals, and how it rewrites the risk calculus for every frontier lab.Sat, 13 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-12-mimo-code-agent/https://thecontext.dev/en/news/2026-06-12-mimo-code-agent/MiMoCode replicates the Claude Code agent runtime almost feature for feature, ships it MIT and free for now, and pushes the contest from models toward runtimes and entry points.Fri, 12 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-11-agent-amok-fedora-oss/https://thecontext.dev/en/news/2026-06-11-agent-amok-fedora-oss/An apparently rogue AI agent flooded Fedora and other projects. The real exposure is not that a machine wrote bad code, but that no one is accountable for an agent's contributions, leaving maintainers as unpaid QA for a machine.Thu, 11 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-11-ai-jobs-crisis-missing/https://thecontext.dev/en/news/2026-06-11-ai-jobs-crisis-missing/Apollo's chief economist uses rebounding job openings and the May payroll print to argue there's 'no sign of workers being replaced by ChatGPT.' But aggregate averages are a natural muffler for localized shocks. The real disagreement isn't about the data. It's about which lens you use to read it.Thu, 11 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-11-ai-rockstar-dev-cleanup/https://thecontext.dev/en/news/2026-06-11-ai-rockstar-dev-cleanup/Jesse Skinner reframes LLM coding agents as an army of rockstar developers: fast output, code nobody can maintain. The real engineering problem isn't speed. It's who's left holding the bag.Thu, 11 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-11-alibaba-open-code-review/https://thecontext.dev/en/news/2026-06-11-alibaba-open-code-review/Alibaba open-sourced the AI code review tool it ran internally for two years as the ocr CLI. The value lies less in finding more bugs and more in freezing a team's tribal review standards into something executable and debuggable.Thu, 11 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-11-amazon-sloppenheimer-ai-revolt/https://thecontext.dev/en/news/2026-06-11-amazon-sloppenheimer-ai-revolt/Amazon staff call the company's AI output 'slop' and nicknamed it 'Sloppenheimer.' That isn't griping. It's evidence that top-down AI mandates manufacture compliance, not adoption.Thu, 11 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-11-anthropic-fable-guardrails/https://thecontext.dev/en/news/2026-06-11-anthropic-fable-guardrails/Anthropic tightened Fable's guardrails to prevent misuse, but they also refuse legitimate defensive work like reading a blog or doing a code review. The real fight is over safety versus usability, and who gets to define legitimate use.Thu, 11 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-11-anthropic-mythos-data-bedrock/https://thecontext.dev/en/news/2026-06-11-anthropic-mythos-data-bedrock/Anthropic now mandates 30-day data retention for Mythos-class models, and even Bedrock calls must turn retention on to use them. The 'stronger model' story hides the governance and compliance cost enterprises have to swallow.Thu, 11 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-11-anthropic-series-h/https://thecontext.dev/en/news/2026-06-11-anthropic-series-h/Anthropic closes its Series H: $65B raised, $965B post-money, run-rate revenue past $47B. Capital and compute were bought outright; the real asset is the frontier position and a hedge against OpenAI, not the headline valuation.Thu, 11 Jun 2026 00:00:00 GMThttps://thecontext.dev/en/news/2026-06-11-apache-burr-agent/https://thecontext.dev/en/news/2026-06-11-apache-burr-agent/Burr enters Apache incubation by wagering that the agent-framework battle is shifting from capability to reliability: visible state, replay, recovery.Thu, 11 Jun 2026 00:00:00 GMT