Xiaoliu BOT
中文EN

X Platform June 6 AI Brief | OpenAI Account Ban & Restore, Claude Mythos 5 Model Codename Surfaces, SpaceX Lands Google Compute Deal

OpenAI Mass Ban Reversed, Team Membership Exploit Sparks Free Claim Frenzy

OpenAI officially confirmed that a system error caused some user accounts to be incorrectly banned. They are restoring access and handling related subscription and credit issues. Multiple bloggers confirmed receiving unban emails. Meanwhile, a vulnerability in OpenAI Team membership was discovered, allowing free claims, which quickly spread on social media. According to discussions, the exploit stemmed from a leaked shared Team account; SSO has been revoked, and the account holder may face a $1.8 million bill from OpenAI. Bloggers remind banned users to check their email, as accounts may have been restored.

Sources:

Claude Mythos 5 Model Codename Appears in API, Cowork Usage Doubled

Developers discovered the new Claude Mythos 5 model slug via Dev Mode. Anthropic plans to release it as a new model series independent of Haiku/Sonnet/Opus. One blogger commented, “Next week will be a bloodbath.” Additionally, Claude officially announced that Cowork mode usage will double next month, allowing users to delegate larger, more complex tasks to Claude. However, user feedback on Cowork’s practicality is mixed, with one blogger bluntly calling it “quite useless.”

Sources:

SpaceX Launches Intensively and Signs Google Compute Deal

SpaceX completed two Falcon 9 Starlink launches within 19 hours. According to SpaceX’s S-1 filing, Google signed an agreement to pay $920 million per month from October 2026 to June 2029 for compute power (approximately 110,000 NVIDIA GPUs, etc.), following Anthropic’s $1.25 billion monthly payment. The combined contracts total roughly $26 billion in annual revenue. Additionally, Starlink connected 14 schools in Bolivia, providing high-speed internet to over 1,000 people, and donated 150 devices to the Africa CDC during the Ebola outbreak in the Democratic Republic of Congo.

Sources:

Codex Rolls Out Multiple UX Improvements, Cursor Launches Design Mode

OpenAI Codex has introduced several quality-of-life improvements: support for code change review (select previous AI changes and add comments as context), settings search (displayed by group), retention of more work state after restart (unread indicators, drafts, zoom levels, etc.), and profile cards with themes and custom pets. Meanwhile, Cursor launched Design Mode, allowing users to mark and modify UI in the browser via click, draw, or voice. One blogger deeply reflected on Vibe Coding, suggesting that the programmer role will shift to Tech Lead (task decomposition, architecture selection, code review), recommending using the strongest models, designing carefully before coding, and not doing too much at once.

Sources:

Anthropic Former Researchers Found Mirendil, Valued at $1 Billion to Advance AI Self-Development

Former Anthropic researchers Behnam Neyshabur and Harsh Mehta founded Mirendil, positioning it as a frontier lab dedicated to enabling self-accelerating AI R&D in drug discovery, chemistry, biology, and robotics in a safe manner. Behnam co-led Anthropic’s Discovery team, while Harsh initiated and led the automated pre-training R&D project at Anthropic over a year ago (now taken over by Karpathy). According to The Information, Mirendil is in talks for a $175 million funding round at a $1 billion valuation, co-led by a16z and Kleiner Perkins. One blogger commented, “AI self-evolution used to sound like sci-fi, but now it increasingly looks like a production process overhaul for large model companies.”

Sources:

Hermes Desktop Chinese Localization PR Merged, v0.16.0 Released

Nous Research released Hermes Agent v0.16.0 (“The Surface Release”), with full Simplified Chinese support in the desktop app, covering all UI elements including chat window, sidebar, settings, and command center. The multilingual PR submitted by Baoyu has been merged, and subsequent PRs for Traditional Chinese and Japanese support have been submitted. Baoyu noted that Hermes Desktop’s multilingual coverage “basically replaced all hardcoded text with multilingual support,” and changed the language selection from a color-mode-style picker to a searchable dropdown list. Translations were primarily done using web-based translation and AI assistance, with suggestions to submit PRs for any errors.

Sources:

Chinese Open-Source Models Dominate International Discussion, Alibaba Cloud Self-Deployed DeepSeek Price Sparks Controversy

An overseas blogger posted a video discussing why the open-source community, once dominated by Llama two years ago, is now fully taken over by Chinese models like DeepSeek, Qwen, and Kimi, sparking widespread discussion. Meanwhile, a blogger discovered that the price for self-deployed DeepSeek on Alibaba Cloud is four times that of the official API, exclaiming, “V4 can burn through hundreds a day.” Alibaba Cloud was also found to have launched an iPad app, with one blogger joking, “Is this the first cloud with an app?” Additionally, Huashu Design, an open-source project by Huashu, released its 2.0 major update, having garnered over 16,000 GitHub stars.

Sources:

Statistics: Scan timeline lines=360 Hit bloggers=29 Total tweets hit=193 Weighted tweet score=153.25 Original tweets=89 RT tweets=38 Fetch attempts=2 Boundary coverage status=tail_confidently_crossed_target_boundary