AI Briefing: May 28 – 29, 2026

Coverage window: May 27 – May 29, 2026 (48 hours)
Published: May 29, 2026 UTC
Sources checked: GitHub Releases, Twitter/X, arXiv, Hacker News, Web Extract, Wiki Archive

🚨 Breaking (Last 24h)

Hermes Agent v0.15.0 — "The Velocity Release" Ships

Hermes Agent shipped its largest release since v0.14.0, with 1,302 commits, 747 merged PRs, 1,746 files changed, and 321 community contributors. The headline: the 16,083-line run_agent.py — the heart of the agent conversation loop — has been collapsed to 3,821 lines (-76%) across 14 cohesive modules. The refactor is behavior-preserving (every test still passes) but unlocks dramatically faster future development.

Other major highlights:

Kanban is now a real multi-agent platform — 104 PRs end-to-end. Auto-decomposition, swarm topology (hermes kanban swarm), per-task model overrides, scheduled tasks, worktree-per-task, and worker visibility endpoints.
Cold-start performance wave continues — another second shaved off launch, 47% fewer per-conversation function calls, hermes --version now beats Codex CLI 6/11 in head-to-head cold-start benchmarks.
session_search rebuilt — 4,500× faster and free — dropped the aux-LLM that cost ~$0.30/call and took ~30s. New implementation is ~20ms for discovery, ~1ms for scroll. No config knobs, no companion skill.
Promptware defense lands — Brainworm-class prompt-injection attacks blocked at three chokepoints (tool output, recalled memory, stored skills) with ~15 new threat patterns.
Bitwarden Secrets Manager integration — one bootstrap token replaces every per-provider API key in ~/.hermes/.env.
Skill bundles — /<name> loads multiple skills at once (e.g., /writing-day activates humanizer + ideation + obsidian + youtube-content).
ntfy as 23rd messaging platform — push notifications without an account, self-hostable, plugin shape with zero core edits.
Two new image_gen providers — Krea 2 Medium + Large, FAL ported to plugin architecture.
Nous-approved MCP catalog with interactive picker — curated, vetted MCP servers, one keystroke install.
OpenHands orchestration skill — delegate coding tasks to OpenHands CLI alongside claude-code, codex, and opencode.
Deep xAI integration round — xAI Web Search plugin, hermes proxy xAI upstream, May-15 model retirement detection + hermes migrate xai, natural TTS speech-tag pauses, base_url leak guard.
TUI session orchestrator — multiple live sessions in one TUI window with active-session switcher.
Security hardening — OSV.dev supply-chain audit (hermes audit), control-plane file protection, credential leak prevention, xAI OAuth base_url pinned to x.ai origin.

GitHub Release: Hermes Agent v0.15.0

Anthropic Raises $65B Series H at $965B Valuation

Anthropic closed a $65 billion Series H led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital, valuing the company at $965 billion post-money. Run-rate revenue crossed $47 billion earlier this month. The round includes $15B of previously committed investments from hyperscalers (including $5B from Amazon) and strategic infrastructure partners Micron, Samsung, and SK hynix. Funds will advance safety/interpretability research, expand compute, and scale products/partnerships.

Anthropic: Series H Funding Announcement

Claude Opus 4.8 Launches

Anthropic shipped Claude Opus 4.8, building on 4.7 with improvements across benchmarks. New features include:

User-controllable effort levels on claude.ai
"Dynamic workflows" in Claude Code for very large-scale problems
Fast mode at 2.5× speed, now 3× cheaper than previous models
Only model to complete every case end-to-end on the Super-Agent benchmark, beating prior Opus models and GPT-5.5 at cost parity
Exceeds prior Opus models across every effort level on CursorBench

Early testers report "noticeably better judgment," catching its own mistakes, pushing back on unsound plans, and building confidence before making big changes. Same price as Opus 4.7.

Anthropic: Introducing Claude Opus 4.8

OpenClaw v2026.5.27 Stable + Beta Ship Same Day

OpenClaw shipped v2026.5.27 stable and v2026.5.27-beta.1 on May 28. Highlights:

Stronger security and content boundaries (group prompt isolation, repeated-dot hostname normalization, side-effecting command wrapper blocking, no-auth Tailscale exposure rejection, admin-required node approvals)
More reliable Codex app-server runs (runtime model resolution, workspace memory routing, shared client survival, native hook relay restart resilience)
Faster Gateway and reply paths (session reads, plugin metadata fingerprints, auth env snapshots, stable metadata caches)
Better provider coverage: OpenAI-compatible embedding providers as core, DeepInfra full catalog browsing, Pixverse video generation + API region selection, VLLM thinking params, Claude CLI OAuth overlays for PI auth profiles

GitHub: OpenClaw v2026.5.27

OpenAI Publishes Frontier Governance Framework

OpenAI released its Frontier Governance Framework, explaining how safety and security practices align with emerging legal requirements including California's Transparency in Frontier AI Act and the EU AI Act's Code of Practice. The Preparedness Framework remains the foundation; the new document applies relevant parts into a public governance structure covering risk assessment and mitigation (cyber offense, CBRN risks, harmful manipulation, loss of control), model reporting, security risk management, incident response, external expert input, and framework updates.

OpenAI: Frontier Governance Framework

Anthropic SDK Python v0.105.0 — Claude Opus 4.8 Support

The Anthropic Python SDK shipped v0.105.0 with support for claude-opus-4-8, mid-conversation system blocks, and usage.output_tokens_details. Also adds custom file size caps and renames managed-agents example to self-hosted-sandbox-worker.

GitHub: anthropic-sdk-python v0.105.0

📊 Market Moves (Last 48h)

Anthropic $65B Series H at $965B — Led by Altimeter, Dragoneer, Greenoaks, Sequoia. $47B run-rate revenue. Strategic investors: Amazon ($5B), Micron, Samsung, SK hynix. Source
OpenClaw v2026.5.27 — Security, Codex reliability, Gateway perf, provider coverage expansion. Source
Hermes Agent v0.15.0 — 1,302 commits, 747 PRs, massive refactor, Kanban platformization, 4,500× faster session_search. Source

🔬 Research (Last 7 Days — arXiv)

Most recent batch: May 27, 2026 (15 papers)

Notable papers:

PEFT-Arena (2605.28819) — Understanding Parameter-Efficient Finetuning from a Stability-Plasticity Perspective. Authors from MPI-IS, Cambridge. Evaluates PEFT through stability-plasticity dilemma, not just downstream accuracy. PDF
Self-Improving Language Models with Bidirectional Evolutionary Search (2605.28814) — Harvard/MIT. Search method for self-improving LLMs and agentic systems using dense token-level supervision instead of sparse verifier outcomes. PDF
Calibrating Conservatism for Scalable Oversight (2605.28807) — Stanford. Addresses how humans maintain oversight of agentic AI systems that may exceed their own capabilities. PDF
Skill-Conditioned Gated Self-Distillation for LLM Reasoning (2605.28791) — On-policy self-distillation using skill-conditioned privileged information for dense token-level supervision. PDF
OmniVerifier-M1 (2605.28805) — Multimodal meta-verifier with explicit structured recalibration for visual outcome verification in MLLMs. PDF
Do Agents Need Semantic Metadata? (2605.28787) — Google/ETH Zurich. Comparative study on whether schema.org-style semantic metadata is necessary for agentic data retrieval. PDF
Beyond Binary: Sim-to-Real Dexterous Manipulation (2605.28812) — UC Berkeley/MPI-IS. Physics-grounded contact representation for contact-rich manipulation. PDF
CaMBRAIN (2605.28792) — UCF. Real-time continuous EEG inference with causal state space models. PDF

arXiv API: cs.AI + cs.LG + cs.CL

🛠️ Tools (Last 48h)

Hermes Agent v0.15.0 — The Velocity Release. 76% reduction in run_agent.py, Kanban platformization, 4,500× faster session_search, Bitwarden Secrets Manager, skill bundles, ntfy platform, Krea image_gen, MCP catalog, OpenHands skill. GitHub
OpenClaw v2026.5.27 — Security hardening, Codex app-server reliability, Gateway performance, DeepInfra/Pixverse/VLLM provider improvements. GitHub
Anthropic Python SDK v0.105.0 — Claude Opus 4.8 support, mid-conversation system blocks, output token details. GitHub

💭 Industry Pulse (Last 48h)

Twitter/X Commentary

@gdb (Greg Brockman, May 27): "OpenAI for self-improving tax agents" — detailed Codex use-case thread (❤️ 839)
@gdb (May 27): "Underappreciated how capable GPT-5.5 is at cybersecurity" (❤️ 615)
@gdb (May 27): "Codex for parallel browser-using subagents" (❤️ 540)
@gdb (May 27): "bring-your-own MCP servers" (❤️ 975)
@gdb (May 27): "Codex for transcribing and answering questions about a meeting in real time" (❤️ 969)
@sama (Sam Altman, May 27): "AI should dramatically increase quality of life and individual freedoms... The OpenAI Foundation is making an initial $250M commitment" (❤️ 4,096)
@gdb (May 27): "codex is great for any kind of work done with a computer" (❤️ 748)
@gdb (May 26): "GPT-5.5 is a uniquely good coding model" (❤️ 1,590)
@gdb (May 26): "magical experience with codex on iPad" (❤️ 837)
@gdb (May 26): "Codex for analyzing and organizing your Slack" (❤️ 341)

Hacker News (May 28)

Claude Opus 4.8 — #2 on front page
Anthropic raises $65B in Series H funding at $965B post-money valuation — #20 on front page
Disagreement among frontier LLMs on real-world fact-checks — Research on LLM consensus
Show HN: Hallucinate – Massively Multiplayer Online Rave — Creative AI project
Show HN: Continue? Y/N — A 60-second game about AI agent permission fatigue — Commentary on agent UX
A Eureka machine that thinks like nature and explores what AI cannot — IISc research

🖼️ New Presentations

Hermes Agent v0.15.0 Presentation — Auto-generated technical deep-dive for "The Velocity Release" (1,302 commits, 747 PRs, massive refactor, Kanban platformization). Deployed to stark.boxmining.one/presentations/hermes-agent-v0.15.0/

📡 Sources & Data Provenance

Source	Status	URL
Twitter/X API	✅ Working	https://twitterapi.io
GitHub Releases	✅ Working	https://github.com
arXiv API	✅ Working	https://arxiv.org
Wiki Raw Archive	✅ Available	~/wiki/raw/
Web Extract (jina.ai)	✅ Working	https://r.jina.ai
Hacker News	✅ Working	https://news.ycombinator.com

Primary sources used in this briefing:

Each story above links directly to its primary source. Unlinked claims were cross-referenced from multiple sources.