Critical Vulnerability in Meta's AI Support Feature Enables Instagram Account Takeover
A serious security vulnerability in Meta's AI support feature was reported on Hacker News. For Instagram accounts with the AI support option enabled, an attacker connects via VPN from the same region as the target account and asks the AI agent to send a code to an arbitrary email address. Simply returning the received code to the agent triggers the issuance of a password reset link, allowing the account to be completely taken over.
The vulnerability has existed for at least several days, and more than 100 high-value accounts have already been hijacked. The procedure has been widely shared via Telegram within black hat communities, and an urgent response is being demanded. The reporter is calling on Meta to temporarily disable the AI support feature and restore the accounts and usernames taken over in the past few days.
While the move to embed AI agents into customer support is accelerating across companies, this incident has highlighted the risk of AI judgment within authentication flows being exploited as an attack vector.
G7 Agrees on a Common Language for Open Source and Open Weight AI
The G7 has agreed on common definitions and language for open source AI and open weight AI. The story is gaining traction on both Hacker News and Reddit's LocalLLaMA community.
Until now, the term "open source AI" has meant very different things depending on who uses it. The common language agreed by the G7 clearly distinguishes between open weight models and truly open source AI, with the details laid out in a document published by the French government.
Community reactions include sentiments such as "this is nothing we didn't already know, but the fact that understanding has advanced at the government level is significant." It is fair to say that international discussions on AI transparency and reproducibility have finally entered the stage of substantive definition-building.
Erin Brockovich Fights Back Against Data Center Secrecy
Environmental activist Erin Brockovich is turning her attention to the opacity of data centers amid AI's rapid expansion. According to TechCrunch, she has launched a new initiative, citing insufficient disclosure of information about the environmental impact of data centers.
The surge in data center demand driven by the AI boom has long been a concern from the perspectives of power consumption and water resource usage. With Brockovich entering the fray, the debate over the environmental impact of data centers could expand even further.
Fujitsu Acknowledges the Limits of the "Man-Month Model" — A Shift in the SI Business for the AI Era
Fujitsu has announced its "Medium- to Long-Term Management Vision 2035," and President and CEO Takahito Tokita has acknowledged that the conventional "man-month model" has reached its limits. According to ITmedia, he explicitly stated that a transformation away from the labor-intensive SI (system integration) model is now necessary in the AI era.
For the leader of a top industry player to publicly declare the limits of man-month-based pricing — the foundation that has supported the business model of Japanese SIers — is a symbolic event. As AI-driven gains in development efficiency accelerate, a fundamental rethink has become unavoidable to keep the traditional business model sustainable.
The Rising Value of Software "Design" in the Age of AI Coding
On Qiita, a counterargument to the view that "AI writes the code, so design is no longer necessary" is drawing attention. The claim is that now, with AI handling code generation, the importance of architecture design and design patterns is greater than ever.
Indeed, human design judgment becomes even more critical for ensuring the quality of AI-generated code and maintaining the integrity of the system as a whole. The shift in developers' roles from "writing code" to "designing and evaluating" as AI spreads also echoes the SI model transformation that Fujitsu's President Tokita pointed out.