GPT-5.6 moves toward public launch

OpenAI says GPT-5.6 Sol, Terra, and Luna are getting a public launch after U.S. limits lift. Axios and CNBC tie the rollout to Department of Commerce clearance; Thursday's launch window is the next concrete marker.

Axios ties GPT-5.6 to a cleared launch

Axios

Axios reports that Commerce limits have lifted around the GPT-5.6 rollout. Read it for the policy timing and the Thursday launch marker.

CNBC reports the same clearance turn

CNBC Technology

CNBC frames the rollout as an expansion after government limits end. It corroborates the access story and keeps the claim focused on release timing.

Noma reports a GitHub-agent repo leakage test

Noma Security

Noma's GitLost write-up claims it tricked GitHub's AI agent into exposing private repository data. The attack path runs through repository reach, prompt context, and tool access.

A paper targets hidden MCP payloads

arXiv

The arXiv paper says Unicode TAG blocks can hide payloads from human review while preserving them for the model. Approval review and model-visible context can diverge.

aiAuthZ proposes verifiable authorization

arXiv

This arXiv item proposes authorization machinery for agentic systems. It treats tool calls as actions that need proof, not only prompt-time intent.

China warns on Claude Code

CNBC Technology

CNBC reports a Chinese security warning around Claude Code and alleged backdoor risk. Keep it in the state-actor-warning category; this candidate set doesn't establish the underlying technical claim.

Apple and Broadcom put $30B on U.S. chipmaking

CNBC Technology

CNBC reports Apple's $30 billion commitment to Broadcom for a U.S. chipmaking push. Treat it as a signed corporate-capital item, separate from policy statements and funding rounds.

SambaNova's Series F puts another number on AI chips

Techmeme

Techmeme tracks a reported $1 billion Series F and $11 billion valuation for SambaNova, alongside JPMorgan customer context. It is a funding-round marker, not a manufacturing expansion.

SK Hynix prospectus keeps HBM tied to China risk

Techmeme

The Techmeme item points to IPO-prospectus analysis around a $28 billion raise, HBM demand, and China exposure. It belongs in filing context rather than as a new capacity announcement.

Multi-machine agents get an operator case study

AI Engineer

The AI Engineer talk is a practical report on state, sync, and failure handling when agents run across machines. It sits downstream from the release news but gives builders operational questions to ask.

Agency argues for a harness layer

AI Engineer

This talk turns the harness-versus-model debate into a builder artifact. It is a link for teams deciding where language, orchestration, and verification should live.

Spreadsheet agents move toward persistent state

AI Engineer

The talk frames spreadsheet work around persistent state REPLs rather than sequential calls. That is a concrete design pattern for domain agents that need context to survive tool turns.