Today's AI news: ChatGPT accounts hijacked, Google's AI Mode hits revenue, malware ranks victims via LLMs, and Anthropic faces a breach and content row.
Here is what happened in AI today — a busy mix of security incidents, business consequences, and platform controversies.
Researchers Reveal How They Hijacked ChatGPT Accounts
Security researchers at Parameter AI have published a detailed post-mortem explaining how they were able to take over ChatGPT user accounts. The writeup, titled Do the math: How we took over ChatGPT accounts, attracted significant attention on Hacker News with over 270 points, indicating it struck a nerve in the security community. The disclosure underscores ongoing concerns about account security for widely used AI platforms. OpenAI has not yet publicly responded in today's coverage.
Google Admits AI Mode Is Hurting Search Revenue
In what is being described as the worst stock-related development since Google's 2004 IPO, the company has acknowledged that its AI Mode in Search is cannibalising advertising revenue. Reports indicate a 15% drop linked to the shift, contributing to roughly a $500 billion fall in market capitalisation. The Latent Space AINews digest frames this bluntly: you cannot have AI-first search and preserve traditional ad revenue at the same time. Analysts are beginning to ask whether a broader AI bear market could follow if other major platforms face similar structural revenue conflicts.
Malware Is Now Using LLMs to Rank Victims by Profitability
Cisco Talos intelligence researchers have uncovered a sophisticated infostealer — dubbed Volta — that uses a large language model to evaluate and prioritise victims based on how financially valuable they are likely to be. Rather than indiscriminately exfiltrating data from every infected machine, the malware uses AI-driven appraisal to rank targets, making attacks more efficient and harder to detect at scale. The story was among the most-discussed items on Hacker News today and represents a notable escalation in the operational use of AI by malicious actors.
Anthropic Breached by AI Startup; Claude Also Draws Fire Over Reverse-Engineering Blocks
Anthropic is dealing with two separate controversies today. First, reports from India Today reveal that an AI startup was able to breach Anthropic's systems within 12 hours, gaining access to employee details and internal messages — a serious security incident for one of the industry's leading safety-focused labs. Second, and separately, users are reporting that Claude is now refusing to assist with video game cloning and reverse-engineering tasks. Tom's Hardware reports that non-malicious, generic reverse-engineering requests are also being blocked in some cases, raising questions about where Anthropic has drawn the line between intellectual property protection and legitimate developer use.
OpenAI Makes Its Case for Fair Use in AI Training
OpenAI has published a blog post compiling excerpts from court briefs submitted by creators, academics, accessibility advocates, and legal experts — all arguing that fair use doctrine supports the use of copyrighted material in AI training. The post is clearly timed to bolster OpenAI's legal position in ongoing litigation, and it signals that the fair use debate is entering a more public, organised phase. The company is framing broad access to training data not merely as a corporate interest but as a matter of social benefit, with accessibility advocates among those cited in support.
Research: Generative AI Is Self-Reportedly Reducing Critical Thinking
A peer-reviewed paper published in the ACM Digital Library — The Impact of Generative AI on Critical Thinking: Self-Reported (2025) — is gaining traction, with 55 points on Hacker News. The study draws on self-reported data and finds that people perceive their own critical thinking abilities as diminishing with increased AI use. While the self-reported methodology has limitations, the findings are consistent with a growing body of concern about cognitive offloading to AI tools and add academic weight to an increasingly prominent public debate.
That is today's AI news digest. Check back tomorrow for the latest updates from across the industry.
Frequently Asked Questions
How were ChatGPT accounts hacked?
Security researchers at Parameter AI published a detailed account of how they were able to take over ChatGPT user accounts, describing the specific vulnerability they exploited. The disclosure attracted widespread attention in the security community.
Why did Google's share price drop sharply in relation to AI?
Google admitted that its AI Mode in Search is hurting advertising revenue, contributing to a roughly 15% revenue decline and an estimated $500 billion drop in market capitalisation — described as the worst such event since the company's 2004 IPO.
What is the Volta infostealer and how does it use AI?
Volta is a piece of malware uncovered by Cisco Talos researchers that uses a large language model to rank infected victims by their likely financial value, making the attack more targeted and efficient than traditional indiscriminate infostealers.
Why is Anthropic's Claude blocking reverse-engineering requests?
Anthropic's Claude model has begun refusing requests related to video game cloning and reverse-engineering. Users report that even non-malicious, generic reverse-engineering tasks are sometimes blocked, raising concerns about over-restriction of legitimate developer work.