Hindsight Memory-PRM gets a local 8B memory-management policy to 77.5% on LoCoMo, beating its API teacher (65.1%) and Mem0's official setting (74.7%, using 8x the context tokens); Selective Forgetting uses paired bootstrap CIs to show graph-structured memory does not beat matched-budget flat vector retrieval (token F1 0.417 vs 0.468); TRACER uses reinforcement learning to decide per-tool retention ratios, cutting 29-46% of tokens in production without hurting task success
Claude Fable 5.1 quietly topped CursorBench and went GA on Bedrock, but Anthropic still hasn't officially announced it; the Pentagon added ChatGPT and Grok to a military AI platform, reportedly bypassing Anthropic; METR disclosed an API key theft that burned roughly $600,000 in inference credits, while NVIDIA's SkillSpector and AIR Security's $50M in combined Seed rounds both point to agent supply-chain trust becoming the new battleground; South Korea's 'AI for All' program starts beta in September aiming to give 52 million citizens free access to a homegrown AI agent by year-end, while Taiwan's financial sector is still working out AI agent governance and accountability basics
openclaw/openclaw, a self-hosted personal assistant, has climbed to 388k stars by wiring WhatsApp, Telegram, Slack and other chat channels into one Gateway. The same week, NVIDIA shipped SkillSpector, which scans Claude Code, Codex, and MCP skills for 71 vulnerability patterns — research it cites found 26.1% of skills contain vulnerabilities and 5.2% show likely malicious intent. Also today: stablyai/orca turns parallel multi-agent coding into a full IDE, and VectifyAI/PageIndex challenges the assumption that RAG needs a vector database with a reasoning-based tree index. claude-code v2.1.257 adds a Containment Escape security rule, and agno v3.0.5 stops swallowing embedding failures silently and starts reporting them honestly.
ML System Design interviews don't test whether you can draw an architecture diagram — they test whether you can articulate the trade-offs at every layer. Today covers four high-frequency topics: how a feature store guarantees training/serving consistency, how to sequence deployment strategies with automated rollback triggers, why monitoring needs to split into system/data/model layers with PSI/KS tests for drift, and how to budget latency for online inference down to the millisecond. The practice problem is 'design a real-time e-commerce recommendation service,' walking through requirement clarification, scale estimation, feature store design, and deployment strategy end to end.
CursorBench 3.2: Fable 5.1 Max hits 73.4% (previous leader Grok 4.6 Extra High was 70.8%) and debuts by taking both first and second place; Fable 5.1 Max beats the old leader by 2.6 points while costing only $9.64 per task, 44% cheaper than the prior Fable 5 Max at $17.32; Anthropic's own site still lists only Fable 5, with no official Fable 5.1 announcement
Agno 3.0.5 highlights: (1) Knowledge ingestion no longer swallows embedding failures — a new partial status sits between completed and failed, and embedders raise EmbeddingError instead of returning an empty vector; (2) Breaking: code catching ModelProviderError around Bedrock embedding failures stops working — switch to EmbeddingError — and the content status API returns 404 for missing content again; (3) adds an opt-in embedding retry, a GandrTools text-to-speech toolkit, an llmman model provider, and an embed_before_replace guard that stops a failed re-ingest from wiping existing data.
AIR Security exited stealth with two seed rounds totaling $50M — $10M led by Sequoia, then $40M led by Greenoaks. This is VC money betting directly on founder pedigree and timing in the unproven 'AI agent supply-chain security' niche, rather than waiting for a Series-A-grade growth curve first.
Tripo AI (parent company VAST) closed a combined Series B and Series B+ round worth roughly RMB 3 billion (about $420M), led by MPCi with heavy follow-on from gaming and entertainment strategic investors. This is China's market treating 3D-native foundation models as their own infrastructure category worth a big bet, not just a stopgap built by bolting 2D generative models onto 3D.
Gemini Omni 1.1 Flash (gemini-omni-1.1-flash): GA since 2026-08-27, replacing the preview that launched 6/30; closed-source, priced per output second: $0.03 at 360p, $0.10 at 720p (default), $0.15 at 1080p, $0.30 at 4K (1080p/4K are upscaled); ranks #1 on the Artificial Analysis Text-to-Video Arena without audio (1322 Elo) and #2 with audio (1237, trailing Wan3.0's 1241); adds scene extension (10s of context, chainable up to 40s total) and first/last-frame interpolation for camera control
Strategy questions rarely fail because you can't size a market — they fail because you conflate 'this market is worth pursuing' with 'we can actually win it.' Exponent's latest Google Product Strategy Interview Guide includes a classic market-entry question: should Google enter the online furniture retail market. Today we break it down with Porter's Five Forces and TAM-SAM-SOM, using Zoom's fight into a market already locked down by Cisco WebEx and Skype as the case study — strategy questions ultimately hinge not on market size, but on whether you have an edge competitors can't copy.
METR (a nonprofit that evaluates frontier AI models' ability to carry out long-horizon agentic tasks) published a security update on August 31 covering two 2026 incidents. In March, a researcher ran a 'vibe-coded' agent orchestration dashboard on a personal EC2 instance meant to sit behind Google auth; a fail-open bug silently disabled that authentication, exposing the system publicly for several days. Attackers likely found it by scanning certificate transparency logs for newly registered sites with high-signal LLM/agent keywords, then prompted the exposed agent directly to reveal its model-provider API key, added an SSH key for persistence, and used the stolen credential to consume roughly $600,000 worth of inference credits over three weeks (credits the model provider had granted METR for free, so not a direct financial loss to METR). In May, METR was targeted by a likely financially motivated attacker running systematic infrastructure scans and staff phishing; during the same window, a bug in a read-only SQL query mechanism behind METR's public transcript viewer let a database meant to hold only public-model data accidentally include some sensitive model output — caught and patched after an independent researcher responsibly disclosed it, with no evidence attackers ever exploited it. Defense takeaways: treat public-facing agent deployments as production infrastructure, put spend caps and anomaly alerts on every API key, and architecturally isolate public endpoints from internal systems.
mcp-spend-guard is an open-source stdio proxy for MCP that fills the gap MCP has no built-in rate limiting for: spend caps, rate limits, a circuit breaker, and a kill switch. Install: `pipx install .`. It addresses the fact that a looping or prompt-injected agent can hammer a paid tool with nothing to stop it.
K-GAT lets retrieved evidence shape collaboration topology, beating the LLM-Debate baseline by 15.7 points on GPQA at under half the token cost; DoCtOR reflects only the decisive-error agent instead of the whole team, lifting success rates by 22%, 26%, and 27% on three datasets; GOD is a local-first control room for agent societies that recorded 78 of 84 targeted moves correctly, though it is validated only at demo scale under one model configuration
Uber revealed the full picture of its agent software factory: 70% of PRs now come from agents, 3,600+ skills sit in a shared registry, weekly agent requests grew 9.4x while total spend held flat; Visa, Mastercard, and Fiserv joined the 25+-member Agentic Payments Alliance to standardize authorization before agentic commerce hits an estimated $3-5 trillion; Nvidia is investing $3.5B in MediaTek convertible bonds to deepen their edge-to-cloud AI computing partnership; Taiwan's government is budgeting NT$40 billion next year toward training 500,000 AI professionals; OpenClaw 2.0 shipped with 933 contributors and 16,000+ merged PRs, the largest single release in the open-source agent project's history
HKUDS/nanobot hit 47.5k stars in half a year, demonstrating the 'small core + multi-channel + long-term memory' formula for a self-hosted personal agent; zhayujie/CowAgent (formerly chatgpt-on-wechat) reinvents an old chatbot wrapper as a full Agent Harness with a three-tier memory architecture and a nightly 'Deep Dream' distillation pass; conductor-oss/conductor wires a durable-execution graph engine to native MCP tool calls, letting an agent's loop survive a crash or a weeks-long human approval wait; mksglu/context-mode goes straight at the pain point of MCP tool calls flooding the context window, and hit #1 on Hacker News. agno v3.0.4 is the only framework release that clears the bar — it flips KnowledgeManagementTools' ingest_path to opt-in by default to close a security gap.
Deep Learning & NLP interviews don't test whether you can draw a transformer diagram — they test whether you understand what each design decision trades away. Today covers four high-frequency topics: how self-attention computes and what KV cache saves, the vocab-size trade-off in tokenization, the senior-candidate detail that 'an embedding is a vector-space contract, and matching dimensions doesn't mean compatible,' and how to choose between fine-tuning and prompting while guarding against catastrophic forgetting. The practice problem is Scale AI's 'design an embedding and classification API,' walking through version compatibility, partial failure, and multi-tenant isolation end to end.
DeepSeek-V4-Flash-Vision-Exp: 284B total / 13B active-parameter MoE, 1M context, 384K max output; priced identically to plain V4-Flash (Input $0.44 / Output $1.32 at peak, half that off-peak); wins 6 of 7 text-agent benchmarks against its own predecessor (DeepSWE hits 59.3, edging past Opus-4.8's 58.0); multimodal-agent scores close in on Opus-4.8 (trails by 2.9 on ApexBench, actually leads on ZeroBench); each image is capped at 384 tokens / roughly 800×800 resolution, trading fine detail for near-zero cost
Metrics questions rarely fail because you picked the wrong metric — they fail because you can't say why that metric represents user value, or you mistake correlation for causation. Exponent's latest 2026 real-interview roundup includes a Meta-style execution question: comments are up but watch time is down, what do you do. Today we break it down with a metric tree, using Facebook's famous '7 friends in 10 days' north star metric as the case study — it found Facebook's growth lever, and it also became one of Silicon Valley's most-cited correlation-causation traps.
AFP arrested two Western Australian men on Aug 26 accused of leading TeamPCP (the crew behind the Shai-Hulud worm), facing 14 combined charges and up to 20 years. Charging details and independent security research show the attack chain: steal Trivy's publishing credentials, cascade into Checkmarx KICS, then exploit LiteLLM's build pipeline for not pinning Trivy to a verified version — the poisoned Trivy stole LiteLLM's own publishing token, which was used to ship a backdoored release. LiteLLM is an AI gateway that centralizes credentials for multiple LLM providers, so this supply-chain attack reached directly into AI infrastructure. An estimated 1,000+ organizations, 500,000+ credentials, and 300GB of data were exposed, with victims including Mercor, OpenAI, and the European Commission. Mitigations: audit for use of the poisoned Trivy/LiteLLM builds, rotate every exposed credential, and pin all GitHub Actions workflows to verified commit SHAs.
read4all is an MCP server that converts PDFs, Office files, images, and web documents into Markdown plus images, preferring MinerU's cloud engine and falling back to local libraries when no API key is set. Install: uvx read4all. It solves the problem that an agent reading an attachment either loses all layout structure or needs a hand-built conversion pipeline.