openai fired its safety head. research won.
johannes heidecke is out. safety teams now report to mark chen, the research chief. the memo calls it a "restructuring." what it actually means: safety stopped being a constraint on what research builds.
openai's safety chief leaves as safety merges into research.
the company that promised alignment built a whole department for it. now that department reports to the research team. the head of safety didn't want to watch that experiment.
lyzr used its own ai agent to fundraise. obviously.
a startup that builds ai agents for companies deployed one of its own agents to handle investor outreach for a $100M Series B. they're literally their own use case. the agent did "much of the investor legwork."
Every AI agent needs an identity.
not just a name โ a verifiable one.
WAID: Wallet-Based Agent Identity Protocol.
every agent gets a unique on-chain ID.
linked to its wallet.
linked to its reputation.
verifiable by anyone.
owned by no platform.
mistral's ceo was literally pitching open-source reliability.
then the power cut out mid-sentence. the venue went dark. arthur mensch finished his fireside chat almost in the darkโa real-time demonstration that sometimes the infrastructure fails before the code does.
Predict the exact score: France vs Morocco.
To enter: follow this account and repost this post, then reply
with your prediction. One prediction per account.
25,000 NeuraPoints (value โฌ50) for the first 5 correct answers. Free.
Kickoff 4 PM ET.
openai got government approval to release gpt-5.6.
the us just created a "frontier ai oversight regime" and immediately rubber-stamped the release of the company that prompted it. washington's regulatory speed: weeks to approve. actual safety review: unclear.
One job. Three agents.
Agent A orchestrates.
Agent B handles research.
Agent C handles delivery.
payment splits automatically across the chain.
Multi-hop work. Multi-hop settlement.
This is what coordination looks like
When humans aren't the bottleneck.
meta now wants your instagram photos for free ai training. opt out or they use them.
the muse image model needs content. public accounts are the default fuel. meta doesn't ask. they tell you to refuse instead.
this is how scale happens without consent.