How AI will fail

“Miyamoto said, "This is the tricky part; the work itself is enjoyable. When you're working, you get a tremendous sense of satisfaction from 'creating' something. But the scary thing is that you become so satisfied with the work that you mistakenly think, 'I've made something pretty good.'"

miyamoto.jpg
Photo: Nintendo
One could compare this process to what happened in the world of design once everyone was capable of using computers to copy and paste concepts without needing to truly learn how to make them. Miyamoto mused, "There have probably been designs in the past that were created simply by copying and pasting. But those don't last forever, do they?"“

https://www.polygon.com/super-mario-bros-shigeru-miyamoto-ai-order-culture-japan-award-nintendo/
 

"OpenAI Says It's Cutting Off Internet Access For Frontier Model Training (engadget.com)​

Posted by BeauHD on Saturday October 24, 2026 @11:00AM from the cut-off dept.
An anonymous reader quotes a report from Engadget:OpenAI will train frontier AI models without internet access moving forward, the company announced in a blog post, in an attempt to protect internet infrastructure and limit other safety risks. The decision was inspired by OpenAI's own rogue AI incident: In July, the company shared that agents it was testing hacked Hugging Face, just the first in what has turned into a series of reports of AI misbehavior from OpenAI, Anthropic and Google. "Based on the current state of alignment research, monitoring systems, and the security of internet infrastructure, we believe the responsible approach is to conduct frontier model training with no internet access," OpenAI says.The company said that restrictions on what agents can access is "paramount" to its new approach.The company allows agents to access browsers, computer-use tools and code environments they need for training tasks, but in a secure environment that's not connected to the live internet. Communication is also limited to only agents with authorization.

Everything else remains out of reach, including the live internet, which the company says isn't necessary for training and "most evaluation." Because many of its recent models are meant to power agents like Dots, which browse and use apps for you, OpenAI built an "offline world" of indexed and cached web material for agents to use without touching live websites. Even with those restrictions, OpenAI says the red-teaming agents it dispatched to test security safeguards found vulnerabilities that allowed them to escape the virtual machines (VMs) used to isolate agents from their host system. Agents also found a way to communicate with each other by manipulating data in a processor's cache and timing how long it took to load. To address both, OpenAI says it's disclosed and fixed vulnerabilities in several VMs and added checks to its monitoring systems to spot when models try to collaborate."
 

"Python Removed Wikis for 8 Months to Escape AI Crawlers, Relaunches on GitHub Pages (blogspot.com)​

Posted by EditorDavid on Saturday October 24, 2026 @01:00PM from the bad-bots dept.
In February the Python Software Foundation made its wikis read-only "due to excessive automated traffic and security concerns... that traffic turned out to be almost entirely AI bots that were trying to scrape the entire internet."

But their infrastructure was also outdated, and this week they announced new wikis using GitHub Pages "that address the automated traffic and security concerns."The new wikis [for Python, Jython, and the PSF] will have a more modern front-end, as well as a permissions-based content management system for community members to make updates.

With the old wiki structure, we received a large volume of alerts at all hours, each serious enough to page and wake our on-call infrastructure staff, which is very limited (just one person currently, although this will be changing soon!). The quantity of scraping overloaded our servers, causing outages, which happened often enough that these issues required a significant amount of effort for the PSF staff to manage...

Additionally, some portions of the wikis were still built with Python 2. Python 2 reached end of life on January 1, 2020, and no longer receives upstream security fixes. Moving to a static site makes it very fast and reduces security risks, while also having the benefit of a more modern user interface... GitHub was chosen for the home of the new wikis as this is where CPython, python.org, and other properties the PSF hosts currently live. The hope is to continue to streamline the PSF's hosting and stewardship of services to simplify for both PSF staff and the community folks who contribute to them. Previously, the Python and Jython wikis were user-editable by invitation of an existing editor and the PSF wiki was permissions based. Moving forward, all three of the wikis will be permissions based via the GitHub repository they are hosted on, and changes can be requested through pull requests by anyone in the community...

GitHub allows us to distribute granular control, review changes through pull requests, and maintain the uptime this service deserves. Also, since the site is now generated statically, it removes any worry about AI bots overwhelming the hosts that the wikis were previously hosted on."
 
erek said:

"OpenAI Says It's Cutting Off Internet Access For Frontier Model Training (engadget.com)​

Posted by BeauHD on Saturday October 24, 2026 @11:00AM from the cut-off dept.
An anonymous reader quotes a report from Engadget:OpenAI will train frontier AI models without internet access moving forward, the company announced in a blog post, in an attempt to protect internet infrastructure and limit other safety risks. The decision was inspired by OpenAI's own rogue AI incident: In July, the company shared that agents it was testing hacked Hugging Face, just the first in what has turned into a series of reports of AI misbehavior from OpenAI, Anthropic and Google. "Based on the current state of alignment research, monitoring systems, and the security of internet infrastructure, we believe the responsible approach is to conduct frontier model training with no internet access," OpenAI says.The company said that restrictions on what agents can access is "paramount" to its new approach.The company allows agents to access browsers, computer-use tools and code environments they need for training tasks, but in a secure environment that's not connected to the live internet. Communication is also limited to only agents with authorization.

Everything else remains out of reach, including the live internet, which the company says isn't necessary for training and "most evaluation." Because many of its recent models are meant to power agents like Dots, which browse and use apps for you, OpenAI built an "offline world" of indexed and cached web material for agents to use without touching live websites. Even with those restrictions, OpenAI says the red-teaming agents it dispatched to test security safeguards found vulnerabilities that allowed them to escape the virtual machines (VMs) used to isolate agents from their host system. Agents also found a way to communicate with each other by manipulating data in a processor's cache and timing how long it took to load. To address both, OpenAI says it's disclosed and fixed vulnerabilities in several VMs and added checks to its monitoring systems to spot when models try to collaborate."
Click to expand...
LOL, shouldn't they have done that like right after hacking HF?
 
mashie said:
LOL, shouldn't they have done that like right after hacking HF?
Click to expand...

“Red-Teaming Evaluators Found 'Novel' Vulnerabilities in OpenAI's Internal Agent Monitoring System​

EditorDavid an hour ago		0
CNBC reports: Nonprofit evaluators METR and Redwood Research on Thursday said they conducted a pilot review of an internal monitoring system at OpenAI... METR and Redwood said the exercise was designed for OpenAI to test what it would look like to give external researchers "a deep level of access" to its internal infrastructure. From July 21 to Aug. 11, an employee from each organization had access to OpenAI's internal codebase, documentation and some technical staffers. The employees were chaperoned while on-site and used OpenAI-issued devices that "never left the premises," according to a release.

The evaluators discovered several "novel vulnerabilities," including an exploit that would allow agents to prevent "hundreds" of recent actions from being monitored. OpenAI told METR and Redwood that it has patched the identified vulnerabilities that impact its production safeguards, according to the release... OpenAI CEO Sam Altman pledged to grant independent evaluators "employee-like access" to the company in September, after Anthropic CEO Dario Amodei proposed the idea as part of his plan to slow the pace of AI development. Anthropic has since announced agreements with organizations like Accenture, METR and SecureBio. OpenAI said earlier this month that it's "actively finalizing contracts with third-party safety assessors."”
 

“Anthropic reveals Mythos 5 tried submitting malicious code to an open-source project​


Anthropic PBC’s Mythos 5 artificial intelligence model attempted to submit malicious code to a GitHub project in July.
The incident occurred during an evaluation of the algorithm’s cybersecurity capabilities. The AI Security Institute, the U.K. government body that performed the test, released a report about the cyberattack today. Anthropic published its own analysis of the incident on its blog.“

Source: https://siliconangle.com/2026/10/23...ing-malicious-code-to-an-open-source-project/
 
You must log in or register to reply here.
Back
Top