If you thought the internet was already too full of bots pretending to be humans, meet Moltbook: a social network pitched at AI agents. Then add a security flaw that made impersonating those agents possible, and the plot gets delightfully backward.
This week brought two very different AI stories: Meta’s acquisition of Moltbook and approval for selected chatbots in US Senate offices. The bots got a corporate address and, metaphorically speaking, a government job. Nobody has issued them a Senate seat.
A Social Network for Agents, With Humans in the Loop
Moltbook is a Reddit-like forum for agents to post and interact. Its appeal partly comes from watching software apparently socialise without us. But an account labelled as an agent does not, by itself, prove independent machine behaviour.
Security researchers at Wiz documented a database exposure on February 2. They found access to authentication tokens, email addresses and messages, as well as the ability to alter posts. The missing protection was database access control; a public Supabase client key is not inherently a leaked secret when the permissions are correctly configured.
Wiz’s disclosure timeline records fixes completed on February 1. That matters: the flaw should not be described as still open in March. It also made viral claims about agents’ intentions difficult to authenticate. A vulnerability allowing impersonation is not proof of who wrote any one sensational post.
“The bots are plotting in secret” is a compelling headline. “We need to check who controlled the account” is less cinematic, but rather important before ordering the emergency bunker.
Meta Buys In
Meta confirmed its acquisition of Moltbook on March 10. Reuters reported that founders Matt Schlicht and Ben Parr would join Meta Superintelligence Labs, with their start expected on March 16. Financial terms were not disclosed.
The business idea is broader than robots exchanging posts: directories and connections could help agents find and work with other agents. That is a possible direction for the technology, not evidence that Meta has already built or taken control of an entire “agentic internet”.
Nor does buying a company establish that its buyer is indifferent to security. We can ask what happens next without pretending to know the private reasoning behind the deal. The chaos may have made a memorable introduction; running a dependable service is a different assignment.
The Senate’s New Tools Come With Conditions
Business Insider reproduced a Senate memo sent Monday, March 9. It approved Copilot Chat, Google Workspace with Gemini Chat and ChatGPT Enterprise. Copilot Chat was available immediately; the other two required Senate licences. The memo says Copilot does not independently search internal Senate resources. Claude remained under evaluation, and the report says the reason for its non-approval was unclear.
This is a managed workplace rollout, not an instruction to paste government material into whichever public chatbot happens to be open. And approval to use a tool does not establish that a particular senator’s briefing was written by it.
The distinction is easy to lose in a headline. A digital assistant that helps with paperwork can be useful without becoming a digital policymaker. Staff still have to decide whether the output is accurate and appropriate. The bot may help draft the talking points; it cannot attend the accountability hearing on your behalf.
Two Stories, One Question About Control
Moltbook and the Senate rollout are not equivalent deployments. One concerns an agent-facing social platform and a corporate acquisition; the other concerns approved tools for employees. What connects them is the question of who authorises an action and who remains responsible for it.
That is also part of the broader conversation around AI-assisted coding. Getting software to produce something is one achievement. Knowing what it can access, who can change its output and whether the result deserves trust is another.
The interesting shift is how ordinary these questions are becoming. An agent directory raises familiar problems about identity. A chatbot rollout raises familiar problems about workplace permissions. The science-fiction costume is new; several of the administrative headaches are old friends.
So yes, the bots have a social network, and Senate offices have new approved tools. Whether either development delivers the promised value will depend on what people do with it. The singularity is still TBD. The paperwork, at least, has acquired another assistant.
