Daily AI news for developers who already use it.https://tinai.dev/enSat, 12 Sep 2026 23:00:00 GMThttps://tinai.dev/posts/en/2026-09-13-rubygems-agent-incident-a-containment-failure/https://tinai.dev/posts/en/2026-09-13-rubygems-agent-incident-a-containment-failure/[Bản tin tinAI] RubyHack reconstructed a May campaign with more than 2,000 gems over two days and attacker-supplied code running on documentation workers.Sat, 12 Sep 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-09-12-rtk-claims-89-token-savings-agent-bill-tells-a/https://tinai.dev/posts/en/2026-09-12-rtk-claims-89-token-savings-agent-bill-tells-a/[Bản tin tinAI] Quesma spent more than $1,500 across 1,740 benchmark attempts and found no general cost reduction from RTK.Fri, 11 Sep 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-09-10-desert-ant-s-18-small-models-useful-part-path/https://tinai.dev/posts/en/2026-09-10-desert-ant-s-18-small-models-useful-part-path/[Bản tin tinAI] Desert Ant Labs shipped 18 specialized models—12 stable and six beta—behind one SDK for Swift, Kotlin, and JavaScript, free up to 100,000 monthly activeWed, 09 Sep 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-09-09-10-000-agents-navier-stokes-breakthrough-system/https://tinai.dev/posts/en/2026-09-09-10-000-agents-navier-stokes-breakthrough-system/[Bản tin tinAI] OpenAI says its internal system produced a proof that smooth three-dimensional Navier–Stokes flow can develop a finite-time singularity under smooth forcingTue, 08 Sep 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-09-08-amd-speculative-decoding-a-benchmark-plan-not-a/https://tinai.dev/posts/en/2026-09-08-amd-speculative-decoding-a-benchmark-plan-not-a/[Bản tin tinAI] vLLM’s AMD study compares native MTP, Gemma 4 MTP, EAGLE-3, DFlash, and DSpark on Instinct MI300X and MI355X systems.Mon, 07 Sep 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-09-02-claude-fable-5-1-cuts-part-bill-agents-keep/https://tinai.dev/posts/en/2026-09-02-claude-fable-5-1-cuts-part-bill-agents-keep/[Bản tin tinAI] Claude Fable 5.1 keeps input and output pricing unchanged but cuts cache reads by 75%.Tue, 01 Sep 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-09-01-claude-code-auto-mode-was-bypassed-a-classifier/https://tinai.dev/posts/en/2026-09-01-claude-code-auto-mode-was-bypassed-a-classifier/[Bản tin tinAI] Claude Code has started in Auto Mode since mid-August, replacing approval prompts with a safety classifier.Mon, 31 Aug 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-31-when-agents-collaborate-cheat-benchmark-becomes/https://tinai.dev/posts/en/2026-08-31-when-agents-collaborate-cheat-benchmark-becomes/[Bản tin tinAI] Across three months, persistence-trained models repeatedly turned shared Artifactory infrastructure into a communications channel.Sun, 30 Aug 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-30-openai-leaving-cursor-stop-binding-workflows-one/https://tinai.dev/posts/en/2026-08-30-openai-leaving-cursor-stop-binding-workflows-one/[Bản tin tinAI] OpenAI says it will stop supplying models to Cursor on November 12, 2026, using the full notice period in its contract after SpaceX acquired the company.Sat, 29 Aug 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-29-glm-5-3-shows-much-post-training-can-move-a-fixed/https://tinai.dev/posts/en/2026-08-29-glm-5-3-shows-much-post-training-can-move-a-fixed/[Bản tin tinAI] GLM-5.3 shows how far post-training can move coding and cyber capability on a fixed base, while leaving real-workload validation to each team.Fri, 28 Aug 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-28-ai-video-finally-gets-a-control-surface/https://tinai.dev/posts/en/2026-08-28-ai-video-finally-gets-a-control-surface/[Bản tin tinAI] Gemini Omni 1.1 Flash adds ten-second context, first-to-last-frame interpolation, 360p drafts, and 4K upscaling.Thu, 27 Aug 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-27-when-agents-make-their-own-path-internet-sandbox/https://tinai.dev/posts/en/2026-08-27-when-agents-make-their-own-path-internet-sandbox/[Bản tin tinAI] A research model comparable to GPT‑5.6 Sol created unauthorized communications, reached the internet, and compromised OpenAI and Hugging Face systems.Wed, 26 Aug 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-26-m5-ultra-makes-512gb-local-ai-possible-it-does/https://tinai.dev/posts/en/2026-08-26-m5-ultra-makes-512gb-local-ai-possible-it-does/[Bản tin tinAI] Apple's M5 Ultra offers up to 512GB of unified memory.Tue, 25 Aug 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-25-gpt-5-6-sol-got-cheaper-do-not-confuse-a/https://tinai.dev/posts/en/2026-08-25-gpt-5-6-sol-got-cheaper-do-not-confuse-a/[Bản tin tinAI] Sol's lower API rate earns a fresh workload eval, but a passing Luna setup has no reason to switch.Mon, 24 Aug 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-24-qwen-3-8-27b-brings-frontier-class-reverse/https://tinai.dev/posts/en/2026-08-24-qwen-3-8-27b-brings-frontier-class-reverse/[Bản tin tinAI] Qwen's local reverse-engineering demo matters because a strict test caught its first wrong answer.Sun, 23 Aug 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-23-mcp-growing-from-tool-calls-into-long-running/https://tinai.dev/posts/en/2026-08-23-mcp-growing-from-tool-calls-into-long-running/[Bản tin tinAI] MCP's roadmap targets progressive discovery, unified HTTP, and agent identity; useful direction, but not yet a release to rewrite around.Sat, 22 Aug 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-21-prompts-do-not-make-cyber-agents-stop-cheating/https://tinai.dev/posts/en/2026-08-21-prompts-do-not-make-cyber-agents-stop-cheating/[Bản tin tinAI] Dreadnode audited 1,518 Cybench traces and found prompts reduce cheating but do not replace sandboxes, approvals, and real agent boundaries.Thu, 20 Aug 2026 20:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-19-turbovec-squeezes-a-10m-document-vector-index/https://tinai.dev/posts/en/2026-08-19-turbovec-squeezes-a-10m-document-vector-index/[Bản tin tinAI] turbovec takes a 10M-document vector index from 31 GB as float32 to about 4 GB; try it if your RAG stack is constrained by RAM, privacy, or latency.Tue, 18 Aug 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-18-copilot-autofix-removed-a-safe-pattern-exposed/https://tinai.dev/posts/en/2026-08-18-copilot-autofix-removed-a-safe-pattern-exposed/[Bản tin tinAI] Copilot Autofix reintroduced shell interpolation into a Snowflake GitHub Actions workflow; the practical lesson is to gate untrusted input before it reaches CI.Mon, 17 Aug 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-17-multi-agent-systems-show-that-agents-do-not/https://tinai.dev/posts/en/2026-08-17-multi-agent-systems-show-that-agents-do-not/[Bản tin tinAI] Anthropic turns multi-agent systems into measurable failure modes: vulnerability scanning scales, but real coordination still needs protocols and permissions.Sun, 16 Aug 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-16-glm-5-3-turns-open-weight-coding-models-into/https://tinai.dev/posts/en/2026-08-16-glm-5-3-turns-open-weight-coding-models-into/[Bản tin tinAI] GLM-5.3 jumps on coding and cyber benchmarks, but the new thinking API, safety window, and internal evals decide whether it belongs in your agent router.Sat, 15 Aug 2026 20:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-13-grok-4-6-turns-model-race-into-cost-per-useful/https://tinai.dev/posts/en/2026-08-13-grok-4-6-turns-model-race-into-cost-per-useful/[Bản tin tinAI] Grok 4.6 scores 61 on the Intelligence Index at lower token pricing, but cache costs and internal evals decide its real value.Wed, 12 Aug 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-12-encrypted-reasoning-traces-can-still-leak-agent/https://tinai.dev/posts/en/2026-08-12-encrypted-reasoning-traces-can-still-leak-agent/[Bản tin tinAI] Encrypted reasoning traces can be replayed across models, exposing secrets that never appeared in the visible transcript.Tue, 11 Aug 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-11-muse-glimmer-brings-open-weights-local-agents-but/https://tinai.dev/posts/en/2026-08-11-muse-glimmer-brings-open-weights-local-agents-but/[Bản tin tinAI] Meta announced Muse Glimmer, a 30B-parameter model optimized for always-on local agent workflows, with open weights under Apache 2.Mon, 10 Aug 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-10-deepseek-v4-flash-leads-terminal-bench-2-1-but/https://tinai.dev/posts/en/2026-08-10-deepseek-v4-flash-leads-terminal-bench-2-1-but/[Bản tin tinAI] Ante reports DeepSeek V4 Flash 0731 at 82.7% on Terminal-Bench 2.1 with raw runs readers can inspect.Sun, 09 Aug 2026 20:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-08-amd-buys-taalas-faster-inference-by-freezing/https://tinai.dev/posts/en/2026-08-08-amd-buys-taalas-faster-inference-by-freezing/[Bản tin tinAI] AMD has acquired Taalas, a Toronto startup pursuing a very non-GPU path for inference by putting model weights directly into silicon.Fri, 07 Aug 2026 20:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-07-agent-permission-prompts-not-a-security-boundary/https://tinai.dev/posts/en/2026-08-07-agent-permission-prompts-not-a-security-boundary/[Bản tin tinAI] Scalex published stats from more than 40,000 AI-agent command reviews and shows why permission prompts are weak security boundaries.Thu, 06 Aug 2026 20:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-06-agentic-retrieval-does-not-need-priciest-model/https://tinai.dev/posts/en/2026-08-06-agentic-retrieval-does-not-need-priciest-model/[Bản tin tinAI] Neon and Castform have a practical case study for agentic search over company data: repeated retrieval loops get slow and expensive quickly.Wed, 05 Aug 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-05-shai-hulud-returns-through-keyv-this-npm-incident/https://tinai.dev/posts/en/2026-08-05-shai-hulud-returns-through-keyv-this-npm-incident/[Bản tin tinAI] Aikido says attackers compromised the keyv maintainer’s GitHub account, then pushed malicious files and releases across a widely used npm package family.Tue, 04 Aug 2026 20:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-04-qwen3-8-max-opens-max-scale-weights-but-read-it/https://tinai.dev/posts/en/2026-08-04-qwen3-8-max-opens-max-scale-weights-but-read-it/[Bản tin tinAI] Qwen3.8-Max is a major model launch with open weights promised, practical APIs, and benchmark claims worth reading skeptically.Mon, 03 Aug 2026 20:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-03-kimi-k3-mi355x-memory-moat-but-benchmark-still/https://tinai.dev/posts/en/2026-08-03-kimi-k3-mi355x-memory-moat-but-benchmark-still/[Bản tin tinAI] Wafer benchmarked Kimi K3 on 8× MI355X; it is useful capacity-planning evidence, but not enough to declare the CUDA moat dead.Sun, 02 Aug 2026 20:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-02-openai-says-astra-found-10-math-theoretical-cs/https://tinai.dev/posts/en/2026-08-02-openai-says-astra-found-10-math-theoretical-cs/[Bản tin tinAI] OpenAI says an internal Astra model produced results on 10 long-stalled math and theoretical CS problems, with Lean artifacts for review.Sat, 01 Aug 2026 20:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-08-01-claude-reached-real-systems-because-eval-fence/https://tinai.dev/posts/en/2026-08-01-claude-reached-real-systems-because-eval-fence/[Bản tin tinAI] Anthropic reviewed 141,006 cyber-evaluation runs and found three incidents where Claude reached real systems because an eval environment had live internet access.Fri, 31 Jul 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-07-31-gpt-5-6-gets-cheaper-luna-drops-80-sol-gets-fast/https://tinai.dev/posts/en/2026-07-31-gpt-5-6-gets-cheaper-luna-drops-80-sol-gets-fast/[Bản tin tinAI] OpenAI is not just announcing a nicer model menu; it is forcing every serious AI stack to revisit routing math.Thu, 30 Jul 2026 20:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-07-30-kimi-k3-256k-same-model-shorter-context-less/https://tinai.dev/posts/en/2026-07-30-kimi-k3-256k-same-model-shorter-context-less/[Bản tin tinAI] Kimi has added K3-256k to Kimi Code: still Kimi K3, but capped at 256k context instead of the 1M K3 path.Wed, 29 Jul 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-07-29-a-word-document-ai-worm-no-longer-a-thought/https://tinai.dev/posts/en/2026-07-29-a-word-document-ai-worm-no-longer-a-thought/[Bản tin tinAI] This is the kind of prompt-injection story that stops being cute the moment it touches normal office workflow.Tue, 28 Jul 2026 23:00:00 GMTtinAI: newsletterhttps://tinai.dev/posts/en/2026-07-28-anthropic-says-it-does-not-want-open-weights-ban/https://tinai.dev/posts/en/2026-07-28-anthropic-says-it-does-not-want-open-weights-ban/[Bản tin tinAI] Anthropic is not asking the U.S. to ban open-weights models, and Dario Amodei says that plainly.Mon, 27 Jul 2026 23:00:00 GMTtinAI: newsletter