🤖 AI Briefing: 2026-05-13

Generated: 2026-05-13 00:01 UTC | Coverage: May 11–13, 2026

🚨 Breaking (Last 24h)

OpenClaw Ships Massive v2026.5.12-beta.3 Release

The OpenClaw project dropped its largest beta release of the year — v2026.5.12-beta.3 — with ~317 changes and fixes spanning the entire platform. Published May 12 at 23:38 UTC, this is a substantial infrastructure and feature update.

Key highlights:

Gateway/OpenAI HTTP: Now honors max_completion_tokens and max_tokens on inbound /v1/chat/completions requests, forwarding client token caps to upstream providers.
Per-sender tool policies: Operators can now restrict dangerous tools by requester identity across global, agent, group, core, bundled, and plugin tool surfaces — a major security enhancement.
ACP session lineage: Gateway session metadata is now exposed through ACP listings, enabling clients to render subagent graphs without private side channels.
Cron job inspection: New cron.get support and openclaw cron get <id> CLI command for inspecting stored cron jobs.
iMessage channel filtering: openclaw channels status --channel <name> filtering plus documented BlueBubbles-to-iMessage cutover path.
Fal image editing: GPT Image 2 and Nano Banana 2 reference-image edits now route to /edit with proper image_urls arrays and geometry enforcement.
Plugin SDK overhaul: Deprecated legacy public subpaths, removed provider-specific helpers, exposed runtime model metadata to native plugin tool factories.
Discord voice improvements: Realtime voice diagnostics, pure-JS opusscript default decoder, voice.allowedChannels config, and opt-in native @discordjs/opus.
Slack threading & formatting: Major Slack improvements including replyBroadcast, unfurlLinks/unfurlMedia, Block Kit formatting hints, and DM session routing fixes.
Google/Gemini 3.1 migration: Comprehensive normalization of retired Gemini 3 Pro Preview IDs across onboarding, config, provider catalogs, and proxy routes — all now point to google/gemini-3.1-pro-preview.
Codex app-server: Native code-mode-only for harness threads, OAuth reuse for ChatGPT-authenticated runs, and tool execution reporting to diagnostics.
Memory & context: /context map command for treemap visualization, dreaming promotion write capping, symlink rejection in memory paths, and vector recall platform variant loading.
Security: Redacted secret-shaped payloads, admin/write scope requirements for wiki ingest/search, SSRF IPv4 pinning, and exec approval command highlighting.

Source: GitHub: openclaw/openclaw v2026.5.12-beta.3

Anthropic SDK v0.101.0 Adds AWS Client for Claude Platform

Anthropic shipped v0.101.0 of its Python SDK on May 11, introducing an AWS client for Claude Platform on AWS — a significant enterprise infrastructure move.

Source: GitHub: anthropics/anthropic-sdk-python v0.101.0

📊 Market Moves (Last 48h)

OpenAI Launches Deployment Company with $4B+ Investment

On May 11, 2026, OpenAI announced the OpenAI Deployment Company — a new unit backed by over $4 billion to embed "Forward Deployed Engineers" into enterprises. The company acquired Tomoro to gain experienced deployment engineering talent from day one.

This signals OpenAI's strategic pivot from pure model provider to full-stack enterprise AI services — directly competing with systems integrators and consulting firms.

Sources:

OpenAI Parameter Golf: 1000+ Participants, 2000+ Submissions

OpenAI published "What Parameter Golf taught us" on May 12 — a research retrospective on their competition challenging participants to train the best language model fitting in a 16MB artifact training in under 10 minutes.

The competition explored AI-assisted machine learning research, coding agents, and extreme model efficiency. The Parameter Golf GitHub repo is now public.

Source: OpenAI: What Parameter Golf taught us

Anthropic × SpaceX Compute Deal + Higher Claude Limits

On May 6, Anthropic announced:

Higher usage limits for Claude (effective immediately)
A compute partnership with SpaceX for expanded training infrastructure
Claude for Creative Work initiative (April 28)
Enterprise AI services joint venture with Blackstone, Hellman & Friedman, and Goldman Sachs (May 4)

Source: Anthropic: Higher usage limits for Claude and a compute deal with SpaceX

🔬 Research (Last 7 Days)

arXiv published 15 papers on May 11, 2026. Most recent tracking date confirms active indexing.

Notable papers:

ELF: Embedded Language Flows — Keya Hu et al. (MIT, Harvard) — A framework for embedding language model flows into constrained environments.
DECO: Sparse Mixture-of-Experts with Dense-Comparable Performance on End-Side Devices — Chenyang Song et al. (Tsinghua) — MoE architecture achieving dense-model performance on edge devices, critical for mobile AI deployment.
Dynamic Skill Lifecycle Management for Agentic Reinforcement Learning — Junhao Shen et al. — Framework for managing skill creation, refinement, and retirement in autonomous agents.
Shepherd: A Runtime Substrate Empowering Meta-Agents with a Formalized Execution Trace — Simon Yu et al. — Runtime infrastructure for meta-agent observability and formal execution tracking.
WildClawBench: A Benchmark for Real-World, Long-Horizon Agent Evaluation — Shuangrui Ding et al. — New benchmark focusing on long-horizon real-world agent tasks, addressing gaps in current eval suites.
Equivariant Reinforcement Learning for Clifford Quantum Circuit Synthesis — Richie Yeung et al. — RL approach for quantum circuit synthesis using equivariant neural networks.
Quantifying Concentration Phenomena of Mean-Field Transformers in the Low-Temperature Regime — Albert Alcalde et al. — Theoretical analysis of transformer behavior in low-temperature training regimes.

Source: arXiv cs.AI/cs.LG/cs.CL feed

🛠️ Tools (Last 48h)

OpenClaw v2026.5.12-beta.3 — Platform-Wide Update

As detailed in Breaking above, this release includes:

65+ changes and 252+ fixes
pnpm 11 workspace upgrade
Major dependency refreshes (ACPX 0.33.1, Codex ACP 0.14.0, Baileys 7.0.0-rc10, Google GenAI 2.0.1, OpenAI 6.37.0)
Plugin SDK restructuring with deprecated subpath cleanup
Native app protocol compatibility ranges for v3/v4 gateways
Enhanced Control UI with recovery panels and performance improvements

Source: GitHub Release

Anthropic Python SDK v0.101.0

AWS client for Claude Platform on AWS
Updated examples to Claude Sonnet 4-5-20250929
Migration from Poetry to uv in example shebangs

Source: GitHub Release

💭 Industry Pulse (Last 48h)

Key Twitter Commentary

@sama (Sam Altman, OpenAI CEO) — May 11:

"speaking of things that have gotten over a threshold for me, the combo of the new ChatGPT model, personality, and person..."

Altman also teased whether ChatGPT is becoming a "superapp" and posted a hiring call for "something new."

@gdb (Greg Brockman, OpenAI President) — May 12:

"AI for helping you build apps powered by AI" "Having an agent in your meeting is such a futuristic experience" "Daybreak: our umbrella effort for defensive acceleration, equipping cyber defenders with the best possible frontier AI"

Brockman is actively promoting AI agents for meetings and cybersecurity defense.

@AndrewYNg (Andrew Ng, AI educator) — May 12:

"There will be no AI jobpocalypse. The story that AI will lead to massive unemployment is stoking unnecessary fear."

Ng pushed back against AI unemployment narratives, calling them fear-mongering.

@ylecun (Yann LeCun, Meta AI Chief) — May 12: Continued political/tech commentary, including a pointed response to Noah Smith: "And Mexico will pay for it." — suggesting skepticism about AI policy promises.

@karpathy (Andrej Karpathy, AI researcher) — May 11:

"This works really well btw, at the end of your query ask your LLM to 'structure your response as HTML', then view the ge..."

Karpathy shared a tip about prompting LLMs to output HTML for better structured responses.

@MindTheGapMTG on Anthropic's strategy:

"$300M for an SDK company. Anthropic is telling you models commoditize but the integration layer doesn't."

This commentary references Anthropic's enterprise focus and SDK investments.

🖼️ New Presentations

No new major version presentations triggered in this briefing cycle.

📡 Sources & Data Provenance

Source	Status	Notes
GitHub API	✅ Working	OpenClaw, Anthropic SDK, Hermes Agent releases fetched successfully
arXiv API	✅ Working	15 papers from May 11; most recent date confirmed
Twitter/X API	✅ Working	Key researcher accounts queried; AI search active
Web Extract	✅ Working	OpenAI newsroom and Anthropic newsroom extracted
Web Search	✅ Working	Multiple queries returned relevant results
Wiki Raw Archive	✅ Available	Previous ingestions from May 5–11 accessible

Each story links directly to its primary source. Claims were cross-referenced from multiple sources where available.