OpenAI’s AI agent dots can use CAPTCHAs to log in on users’ behalf
要闻
James Sun of OpenAI’s product team announced that the agent product dot can now use login codes sent to a connected inbox to sign users in, with options to allow one-time use or ongoing autofill.
James Sun, a member of OpenAI’s product team, announced on X that OpenAI’s always-on agent product dot has been upgraded to use login verification codes sent to a connected inbox and complete sign-ins on a user’s behalf. According to his explanation, when a login prompt appears, users can authorize dot to use the code only once or choose to always allow dot to autofill it. James Sun also said that more login improvements are planned for dot.
OpenCode launches Browse, allowing AI agents to control your real browser
开发生态
OpenCode has launched Browse, an MCP server and Chromium extension that let AI agents operate a user’s real browser with existing tabs, logins, and extensions; when login, 2FA, or payment is required, the agent hands the page back to the user and resumes after the user clicks Continue.
OpenCode’s Browse enables AI agents to control a real browser through an MCP server and Chromium extension, using the browser’s existing tabs, login states, and extensions. According to OpenCode’s product page, the agent opens and drives tabs with Playwright, runs them in grouped background tabs, takes screenshots, and records the screen. When a login, 2FA prompt, or payment requires user input, the tab is handed back to the user, and the agent continues after the user clicks Continue. Installation starts with the Chrome extension, which guides users through the remaining steps. The product page also lists a method for installing the CLI directly, while stating that this is generally unnecessary.
OpenCode founder responds to model slowdowns: “We’re adding capacity”
开发生态
OpenCode founder Frank Wang said the platform is adding substantial capacity after users reported that DeepSeek V4.1 Flash was responding slowly and struggling with basic tasks, with conditions expected to improve gradually.
OpenCode founder Frank Wang said the platform is adding substantial capacity to address slow responses from DeepSeek V4.1 Flash, and that conditions are expected to improve gradually. Users had reported that the model was running slowly on OpenCode and that even basic tasks were difficult to complete.
OpenCode co-founder dax said that business growth has made it difficult for him to keep up with every matter. Users who encounter issues related to inference or Go can contact Frank Wang, who is more focused on handling those areas.
Cua and Scale AI launch Cua Driver Bench to evaluate agents’ ability to perform desktop tasks in the background
技术与洞察
Cua and Scale AI’s AgentEnv launched Cua Driver Bench to test whether an Agent can complete macOS desktop tasks in the background while the user keeps working; the benchmark has 10 tasks, and Cua Driver posted a 58% full-pass rate.
Cua announced Cua Driver Bench with Scale AI’s AgentEnv as a benchmark for evaluating whether the computer-use tool layer can execute macOS desktop tasks in the background. The benchmark currently covers only macOS and uses the macOS 26 desktop, with 10 tasks divided into multi-application workflows, precise interaction, and interruption and recovery.
The full-pass rates were 58% for Cua Driver, 33% for Codex computer use, 20% for Vercel’s agent-desktop, and 3% for browser-use. The main measure is background completion: the test simulates a user window remaining in the foreground while the user continues typing, and any one of eight forms of interference, including focus loss, window obstruction, and clipboard changes, causes that trial not to count as completed in the background.
Anthropic must face a lawsuit alleging misleading advertising for Claude Max
行业动态
Alex Tse, a magistrate judge in the U.S. District Court for the Northern District of California, ruled that Anthropic must face claims that its Claude Max usage claims misled consumers. Some claims were dismissed, but the plaintiff may amend the complaint.
Alex Tse, a magistrate judge in the U.S. District Court for the Northern District of California, ruled that Anthropic must face allegations that its Claude Max claim of “5x or 20x more usage than Pro” was misleading. Claude Max is billed monthly at $100 or $200, but the advertised multiples were calculated per 5-hour session, while the weekly usage caps did not increase by the same multiples. Using Sonnet 4 as an example, the plaintiff calculated that the weekly usage of the two Max plans was 3.5x and 6x that of Pro, respectively. The court found that consumers paying monthly could reasonably understand the claim as referring to monthly usage, and that information explaining the session-based calculation was two clicks beyond the sales page, which was insufficient to dismiss the case at this stage. Claims including negligent misrepresentation and breach of contract were allowed to continue. Claims involving California unfair competition, false advertising, and the implied covenant of good faith and fair dealing were dismissed, with the plaintiff allowed to amend and refile the complaint.
OpenAI CFO says the company is still considering an IPO in 2027
行业动态
OpenAI CFO Sarah Friar said in a CNBC interview on November 6, 2026, that the company is still considering a 2027 public listing, without announcing a confirmed timetable.
OpenAI CFO Sarah Friar discussed the company’s listing plans on a CNBC program on November 6, 2026. According to the report, OpenAI continues to consider an initial public offering (IPO) in 2027, but no public information indicates that the company has set a listing date or completed related arrangements.
Anthropic’s IPO valuation comes under scrutiny as it prepares to begin its IPO roadshow next week
前瞻与传闻
Anthropic still plans to begin its IPO roadshow next week and seek a Nasdaq listing before Thanksgiving in the week of November 16 under ANT, while potential investors view the discussed $1.5 trillion valuation as too high and see $1 trillion to $1.25 trillion as more likely.
Anthropic still plans to begin its IPO roadshow next week and seek a Nasdaq listing before Thanksgiving in the week of November 16, with ANT as its ticker. According to The Information, the company had previously sought a valuation of up to $2 trillion. Reuters and other media reported that some potential investors consider the recently discussed $1.5 trillion valuation to be near the high end of expectations, meaning Anthropic may need to offer a larger discount to secure subscriptions. Anthropic declined to comment on the reports.
CNBC, citing investors, reported that a valuation of $1 trillion to $1.25 trillion may be a more realistic target. The Wall Street Journal reported that some investors believe the valuation could be closer to $1 trillion. Investor caution is linked to a broad sell-off in technology stocks, quarterly losses and commitments to substantial AI infrastructure spending. Investors are also waiting for an updated S-1 filing containing the October revenue data that was left blank.
