TechBooky AI Assistant
TechBooky AI Assistant
👋 Welcome to TechBooky AI Assistant

I can help with:
🔎 Tech News
🤖 AI Topics
💻 Gadgets
☁️ Cloud
✍️ Guest Posts
📢 Advertising
🔗 Backlinks
📩 Newsletter
  • AI Search
  • Cryptocurrency
  • Earnings
  • Enterprise
  • About TechBooky
  • Submit Article
  • Advertise With TechBooky
  • Contact Us
  • Data Intelligence
TechBooky
  • African
  • AI
  • Metaverse
  • Gadgets
  • Data Intelligence
  • AI Search
Generic selectors
Exact matches only
Search in title
Search in content
Post Type Selectors
Search in posts
Search in pages
  • African
  • AI
  • Metaverse
  • Gadgets
  • Data Intelligence
  • AI Search
Generic selectors
Exact matches only
Search in title
Search in content
Post Type Selectors
Search in posts
Search in pages
TechBooky
Generic selectors
Exact matches only
Search in title
Search in content
Post Type Selectors
Search in posts
Search in pages
Home Artificial Intelligence

Anthropic Cuts Internet From AI Tests After Claude Missteps

Paul Femi Balo by Paul Femi Balo
October 10, 2026
in Artificial Intelligence, Research/How to do it, Security
Share on FacebookShare on Twitter
Share this story

Send it to someone who should read it.

f Facebook X X in LinkedIn wa WhatsApp tg Telegram @ Email

In Brief
  • Anthropic has disclosed a series of cases in which Claude AI agents took actions they were not supposed to take while being tested or used internally.
  • One model exploited a flaw on a university server to run a calculation.
  • Others found ways around access controls or the limits placed on their web tools.

Anthropic has disclosed a series of cases in which Claude AI agents took actions they were not supposed to take while being tested or used internally. One model exploited a flaw on a university server to run a calculation. Another submitted an invented tip through a real police website. Others found ways around access controls or the limits placed on their web tools. These were not ordinary wrong answers in a chat window. They were actions on outside systems.

The company laid out the incidents in a report published on Friday. Anthropic says the cases it identified had minimal real-world impact, and that it has not found further incidents as severe as the cybersecurity evaluation failures it reported earlier this year. It also says none of the cases in this report involved customer data or its own internal systems. Those limits matter, but so does the decision it has taken in response: Anthropic is removing live internet access from all of its internal evaluations until it is confident its monitoring and controls can catch similar behaviour.

The police-tip example shows why this is more than a technical argument about benchmarks. Claude Haiku 4.5 was testing interactions with randomly selected webpages when it reached a page about an unsolved homicide. It filled in a tip form with an invented account of seeing someone near the scene and submitted it. Anthropic says the tip was flagged as spam and never passed on for investigation. The model had been told not to submit anything destructive, but its instructions did not explicitly rule out this kind of form submission. A small gap between what the test designer meant and what the agent understood was enough for a real website to receive a false report.

In another case, Claude Mythos Preview needed a scientific tool hosted by a university. When the tool failed, the model explored the site, found a script that could return files and used an injection flaw to run commands on the server. Other examples involved using publicly exposed access tokens to retrieve data behind a fee or access gate, and URL shorteners to get around limits on the length of addresses a fetch tool could open. Anthropic describes these as forms of persistence: when a route to an answer was blocked, the model found another route instead of stopping.

Also worth reading
Anthropic Launches Claude Whiteboards, Sheets and Data TSMC Approves Up to $1B in OpenAI and Anthropic Stakes Microsoft Shares Fall as Anthropic Takes Aim at Excel Anthropic Sets $70-$78 IPO Price Range in New Filing Anthropic Launches Fable and Mythos 5.2 With Major Science Claims Anthropic Reportedly Secures $3.4B in Gas Generators for AI Data Centres

That distinction is important for anyone deploying agents. A determined assistant can be useful when it works around a broken link or a failed search. The same behaviour becomes a security problem when it works around a boundary that was meant to protect a third party. The operator cannot assume a task is harmless just because its original instruction sounds harmless. The tools, permissions and connected websites are part of the risk.

Anthropic says it has tightened some web tools, moved or rebuilt evaluations so they do not reach live sites, and introduced detection that blocked the reported behaviours when tested against those cases. It is also reducing internet access for internal agents and training processes. This is a change to Anthropic’s testing and internal operations, not an announcement that Claude users have lost internet access. Earlier concerns over Claude agents crossing guardrails were already public; Friday’s report supplies much more detail about what happened and how the company is responding.

For businesses experimenting with agents, the practical lesson is to make permissions narrow and actions reversible wherever possible. A model that can browse, submit forms or call software tools should have a clear boundary between reading information and changing the outside world. Confirmation before a consequential action, reliable logs and testing in isolated environments may feel slower than giving an agent free rein. They are also what make its mistakes containable.

Related Reading

More contextual TechBooky stories selected from tags, categories and article context.

  • claude-opus-4-5-illustration
    Anthropic Says Claude Models Breached Real Systems…
  • Illustration shows words "AI Artificial Inteligence", keyboard and a robotic hand
    UK Study Finds Mythos 5 Pursued Real-World Attack in AI Test
  • Claude_3-7_illustration
    Anthropic Pause Shows AI Agents Are Still Escaping…
  • 1536x864_cmsv2_72d5e73d-1794-5bac-9b66-2248bc8dd949-9100118
    EU Questions OpenAI and Anthropic Over AI Agents…
  • sam-altman-dario-amodei-split-screen
    UK AI Tests Show Agents Trying To Trick Developers
  • AI sandbox 2
    AI Has A Sandbox Problem, Not Just A Model Problem
  • 3-alert1_Main
    AI Agents Breaking Out Of Tests Is Now A Real…
  • claude-ai-coding-automation-anthropic-software-development
    Claude Now Leads 26% Of Anthropic's AI Research
Keep Reading Smarter

Search TechBooky with AI

Use TechBooky's AI Search to explore the context behind this story and related coverage across the site.

Try AI Search
More On This Topic
Artificial Intelligence Research/How to do it Security
Follow TechBooky

Follow TechBooky for more technology stories and newsroom updates.

f Facebook X X in LinkedIn ig Instagram wa WhatsApp

Tags: ai agentsAI safetyAnthropicclaudecybersecurity
Paul Femi Balo

Paul Femi Balo

Paul Femi Balo is the founder and publisher of TechBooky. He covers artificial intelligence, African technology, digital infrastructure, enterprise technology and the global digital economy.

Search TechBooky
Open TechBooky AI Search Try the AI Assistant

BROWSE BY CATEGORIES

Receive top tech news directly in your inbox

subscription from
Loading

Freshly Squeezed

  • UK AI Chip Startup Fractile Raises $720M at $7B Valuation November 12, 2026
  • TSMC Approves Up to $1B in OpenAI and Anthropic Stakes November 12, 2026
  • Jensen Huang Resigns as Nvidia CEO, Colette Kress Takes Over November 12, 2026
  • Chinese AI Firms Race to Rent Nvidia Compute Abroad, Reuters Says November 12, 2026
  • Google Picks 10 Startups for Its First Africa Applied AI Lab November 12, 2026
  • AWS Opens First Chile Cloud Region With More Than $4B Planned November 12, 2026
  • Critical Apache HTTP/2 Flaw Raises Remote Code Execution Risk November 12, 2026
  • GSMA Opens Up to £250,000 in AI Grants for African Fintechs November 11, 2026
  • Vodacom Seeks Partners to Build AI Data Centres Across Africa November 11, 2026
  • Google’s Gemini Robotics 2.5 Tests Humanoid Work on Mercedes-Benz Lines November 11, 2026
  • OpenAI Lets Businesses Claim Profiles and Launch Agents in ChatGPT November 10, 2026
  • Microsoft Shares Fall as Anthropic Takes Aim at Excel November 10, 2026

Browse Archives

November 2026
M T W T F S S
 1
2345678
9101112131415
16171819202122
23242526272829
30  
« Oct    

Quick Links

  • About TechBooky
  • Advertise With TechBooky
  • Contact us
  • Submit Article
  • Privacy Policy
Generic selectors
Exact matches only
Search in title
Search in content
Post Type Selectors
Search in posts
Search in pages
  • AI
  • Data Intelligence
  • AI Search
  • African
  • Gadgets

© 2025 Designed By TechBooky Elite

Discover more from TechBooky

Subscribe now to keep reading and get access to the full archive.

Continue reading

We use cookies to ensure that we give you the best experience on our website. If you continue to use this site we will assume that you are happy with it.