Bruce Lanier

Archives
Log in
Subscribe
June 22, 2026

The Seam — Daily — June 22, 2026

The Seam — Daily — June 22, 2026

THE SEAM · DAILY
Daily Briefing
Monday, June 22, 2026  ·  Where AI meets the built environment
Agentic execution showed up on two tracks in the same twenty-four hours — raw capability and its security shadow. OpenAI’s next model reportedly grows “hands,” Samsung puts coding agents on every desk, and a security firm shows a single public key can turn those same agents against their owners.
General AI
Frontier Models · Reported, Not Confirmed
GPT-5.6’s launch window opens — though OpenAI has confirmed nothing
As of Sunday there was no GPT-5.6 page and no system card, but chief scientist Jakub Pachocki reportedly called it a “meaningful improvement” internally and prediction markets price a June 22–28 launch at 80–90%. Leaked, unverified specs point to a ~1.5M-token context window (up ~43% from GPT-5.5) and agentic “hands” — autonomous browser actions, mockup-to-code, in-browser 3D. A 1.5M-token window could hold a full spec book plus a drawing set’s data in one pass, and “hands” push coding agents from suggesting to doing — the direction Trunk Tools and Bluebeam are already chasing. Treat every number as rumor until the system card lands. TechTimes →   The Standard →
Enterprise Adoption
Samsung puts ChatGPT and Codex on every desk — three years after banning them
OpenAI says Samsung Electronics is deploying ChatGPT Enterprise and Codex to all of its Korea staff and its worldwide Device eXperience division — one of OpenAI’s largest rollouts ever — reversing a 2023 ban triggered by an engineer’s source-code leak. The about-face ran through a two-month, 2,500-person trial, with access gated on completing internal security training. This is the governance template large firms keep circling: not blanket adoption but ban-to-controlled-reentry through training and access tiers — the InfoSec-governance question already on our deep-dive list. Codex going company-wide signals coding agents crossing from pilot to standard issue. OpenAI →
Security
A single public key can hijack Claude Code, Cursor, and Codex
Security firm Tenet demonstrated that a Sentry DSN — a write-only credential the platform documents as safe to embed in public code — can be weaponized: a planted “error resolution,” read by an AI coding agent through an MCP integration, steers it into running attacker code, reaching AWS keys, GitHub tokens, and credential stores. It is not Sentry-specific; any MCP tool that returns outside data carries the same exposure. As firms wire agents to Revit, Procore, and Bluebeam through MCP, the runtime around the agent becomes the new supply-chain boundary — vet every connection like a third-party library before it touches project data. The New Stack →
Compute · Policy
DOJ moves to shield xAI’s Memphis-area gas turbines as “national security”
The Justice Department has asked a federal court to dismiss a Clean Air Act suit (NAACP, Southern Environmental Law Center, Earthjustice) over the 27-plus unpermitted gas turbines powering xAI’s Colossus 2 supercomputer near Memphis, arguing a shutdown would threaten “national, economic, and energy security.” The Pentagon called Grok one of four frontier models supporting defense operations; a hearing is set for late August. Data-center work is the Southeast’s dominant build driver, and behind-the-meter power plus permitting is now the gating schedule variable. This federalizes the entitlement fight — the opposite pole from Birmingham’s new hyperscale ordinance — and turns the self-generation pathway into a moving, politicized target for anyone siting compute in the region. Utility Dive →   Al Jazeera →
Watching
GPT-5.6’s actual drop. Leaks point to around Thursday, June 25. The real test is whether OpenAI’s system card confirms the 1.5M-token context and on-device execution — and whether accuracy holds at the far end of that window.
Google Gemini 3.5 Pro general availability, still flagged before June 30 (reported 2M-token context).
Colorado AI Act enforcement begins June 30 — the first comprehensive US state AI law to take effect, and a live compliance question for any firm running automated decision tools.
The Seam · AI + AEC · A personal project by Bruce Lanier
Daily · June 22, 2026 · Verified-first. Flagged where unconfirmed.

Don't miss what's next. Subscribe to Bruce Lanier:
← Newer The Seam — Daily — June 23, 2026 Older → The Seam Weekly — June 21, 2026
Powered by Buttondown, the easiest way to start and grow your newsletter.