Posts

AI news flash: The US government issued an emergency directive requiring Anthropic to remove its new Claude models (Fable 5, Mythos 5) from public access by late Friday. Anthropic stated in their official blog that the directive stemmed from a government misunderstanding of the threat level posed by a ’non-universal jailbreak’ vulnerability, which originated from an anonymous report. Claude currently has approximately 18,900 monthly active users, and Anthropic says it’s actively fixing security issues to restore user access as quickly as possible.

AI News Flash: Amazon CEO Andy Jassy reportedly became the source of security concerns that prompted Anthropic to urgently pull two models offline globally. The Wall Street Journal reported that Jassy shared with Treasury Secretary Scott Bessent and multiple government officials that Amazon researchers using Anthropic’s Claude Fable 5 model had successfully obtained information usable for network attacks. The government subsequently imposed export control bans on both Fable 5 and Mythos 5 models, leading Anthropic to cut off access worldwide. An Amazon spokesperson stated that exchanging views with the government on potential security risks isn’t uncommon, though they wouldn’t disclose discussion details, confirming AWS was also impacted by this takedown.

AI News Flash: KPMG, a major international professional services firm, published a report titled ‘Redefining Excellence in the Agentic AI Era’ in October 2025, but multiple organizations cited in the report denied the accuracy of the descriptions, leading KPMG to ultimately remove the report from its website and launch an internal investigation. The case was exposed by AI detection research organization GPTZero, which pointed out to the Financial Times that the report contained multiple fabrications…

AI News Flash: Multiple state attorneys general launch joint investigation into OpenAI; NY Attorney General issued subpoena last Friday demanding documents covering broad topics including company advertising practices, user engagement and retention strategies, model sycophancy issues, consumer and health data handling, and treatment of minors and the elderly. Which states are involved remains undisclosed…

AI News Flash: Anthropic officially released the first public version of the Mythos model series, Claude Fable 5. The company revealed last month that over 10,000 high-severity or critical vulnerabilities targeting systemically important software were discovered during Mythos testing, sparking widespread debate over whether the model should be publicly deployed. Anthropic decided to release it…

AI News Flash: The IPO market is heating up again, but this round’s main players are no longer the FAANG group that dominated the past decade. Market observers have proposed a new acronym - MANGOS - including Meta (or Microsoft, depending on analyst stance), Anthropic, Nvidia, Google, OpenAI, and SpaceX. Half of these six companies are planning to go public…

AI news flash: Anthropic recently rarelly publicly expressed strong dissatisfaction with the government’s decision. Relevant authorities cited a ‘potential jailbreak vulnerability’ found in its most powerful AI model as reason to pull this commercial product already deployed to hundreds of millions of users. Anthropic immediately published a direct rebuttal on its official blog, stating: ‘We do not believe that discovering a limited-scope potential jailbreak vulnerability should be a sufficient reason to pull a commercial model deployed to hundreds of millions of users…’ This strongly worded statement shows the divergence between Anthropic’s current AI safety assessment standards and government regulatory judgment has reached a public showdown. Notably, this incident highlights the tension between ‘safety warnings’ and ‘over-regulation response’ — Anthropic’s own stance on safety transparency may have become the trigger for regulatory action. Since the original summary only provides this quote, details regarding government agency names, specific vulnerability nature, and scale of removal are limited. Please refer to the original link for more details.

AI News Flash: Google announced this Friday that it is suing a Chinese cybercrime group called ‘Outsider Enterprise,’ demanding the complete takedown of its criminal infrastructure. According to Google’s accusations, the group set up 9,000 fake websites, registered over one million fraudulent domains, and sent 2.5 million phishing text messages to Android users within just two weeks, with 55,000 of those being actively reported by users in May alone…

AI News: Meta’s newly formed Applied AI team, just three months old, is facing a severe internal crisis. The organization of about 6,500 engineers and product managers erupted in a public conflict during an internal livestream this week — someone barged in and, using profanity-laced rants, demanded attendees relay a message to a Meta AI executive, causing the situation to spiral out of control. The incident reflects deeper collective anger…

AI News Flash: GitHub Copilot CLI engineering team recently published a major improvement to the agent delegation mechanism. The core issue: in agent systems, more delegation isn’t always better. Previously, Copilot CLI would sometimes unnecessarily spin up sub-agents to search repositories and wait for results on simple tasks, turning what could be done in one step into three steps, with each handoff adding coordination costs, tool call overhead, and waiting time.