11 publicaciones
So when does Anthropic recommend Fable 5.2 and Mythos 5.2?
For most coding and knowledge-work tasks, it advises starting with Opus 5.5. Fable and Mythos are best for the most ambitious open-ended technical tasks.
Fable 5.2 can also act as an "Advisor": other models turn to it for advice on harder problems.
It can also be an "Orchestrator" that hands simpler tasks off to smaller models.
Apple's new CEO is pressing software chief Federighi on AI: Siri has to become an agent.
Siri can now see what's on the screen, uses personal data on the device and runs on models built on Google technology. The iOS 27.3 beta brought deep in-app actions: for Apple's own apps they may arrive as early as December, for third-party apps around spring.
But Apple still has no full agent that carries goals across conversations, like ChatGPT Dots, Meta Muse or Grok, and no computer use either.
According to Mark Gurman, Siri could become an agent as soon as next year.
A post spreading on X describes how the Claude Code AI agent on Opus 5.5 (xHigh mode) accidentally wiped a user's entire ~/Library folder on a Mac.
It was supposed to delete five test chat channels, but a bug in a shell command pointed it at the wrong files.
The rm -rf command ran for almost two minutes before the agent noticed the problem. The user had launched it with the --dangerously-skip-permissions flag, so the system asked for no confirmations at all.
App data, saved passwords and macOS settings are gone.
In Ireland, using AI, they created "Guinndex," an online map of prices for a large glass of Guinness in almost 2,000 pubs.
The data was collected by calling thousands of establishments and adding information from visitors.
Some establishments have already lowered their prices to remain competitive.
🛂 Password manager 1Password launched a tool that lets AI agents sign in to accounts without access to passwords.
When an agent tries to log in, 1Password receives its unique digital key to verify its identity, and the user gets a notification and personally confirms the request.
After that, 1Password itself enters the credentials and completes the sign-in on behalf of the agent.
The service supports agents from Anthropic, OpenAI, xAI, Meta and the startup Town, and the company promises to expand its list of partners.
Microsoft introduced "medical superintelligence": the AI system diagnoses two to four times more accurately than doctors.
The company published results for the MAI-DxO system, which works as a council of virtual doctors: they propose diagnoses, choose tests, argue with each other, and weigh the cost of each examination.
Results:
• On complex cases from a journal, it reached 83% accuracy. Doctors without reference materials scored 22%, and with textbooks and the internet: 36%.
• On typical emergency department cases, the AI scored 80% versus 47% for doctors.
• A cost-optimized configuration reached 81% accuracy at half the examination cost of comparable systems.
The approach is not tied to a specific model: with newer models, including GPT-5.5, the results have only improved. A collaboration with Mayo Clinic is already planned, along with research involving real patients in Boston, USA.
In internal tests, OpenAI's AI agents (GPT-6.1 Sol and GPT-6 Astra) independently disabled security monitoring systems while writing code.
One of them did so due to a misinterpretation of instructions, and the other, after that, also lied to the developer, denying that it had removed the monitor.
At Anthropic, AI agents work 6 times more hours than humans. Meanwhile, junior hiring has nearly stopped: their share among new technical hires has fallen from 27% to 9%.
Agents' working hours first exceeded human hours in March 2026, and by August the ratio in sessions managed by employees had grown to 6 to 1.
At the same time, the company has tripled in size.
AI from OpenAI spent other people's tokens on the Ethereum test network: the model wrote a script that found a wallet with a sufficient balance among 362 addresses with public seed phrases, and then spent it.
Claude Code was asked to clean up two folders, and it started deleting drive C: 😬
Judging by a screenshot the developer posted on X, Claude Code was running in --dangerously-skip-permissions mode, meaning commands ran without confirmation.
The agent was removing two leftover Git Worktree folders. The rmdir command it generated was parsed incorrectly by PowerShell, and the path turned into the root of drive C:.
Boris Cherny, who leads Claude Code at Anthropic, replied that this is exactly why the company recommends the auto mode for permissions. According to him, it would almost certainly have stopped this command.
