I see this security warning in my log. It’s coming from localhost, which is confusing. If I didn’t know better, I’d say bots are scrubbing the web and happened to come along my account, then tried to logon to “learn” whatever.
I searched the forum and others report this happening once or twice, only for it to stop on its own without explanation. It’s a little disconcerting.
Logger: homeassistant.components.http.ban
Source: components/http/ban.py:138
Integration: HTTP (documentation, issues)
First occurred: August 15, 2026 at 10:21:53 AM (3 occurrences)
Last logged: 10:01:18 AM
Login attempt or request with invalid authentication from localhost (127.0.0.1). Requested URL: '/api/config'. (Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko); compatible; ChatGPT-User/1.0; +https://openai.com/bot)
Login attempt or request with invalid authentication from localhost (127.0.0.1). Requested URL: '/api/config'. (anthropic-ai)
I don’t use either of them, but I think the errors can be attributed to a faulty AI integration, trying to read its own config, and not refreshing credentials properly. You should probably report this to the author.
If you don’t have a local AI, then why did you think one was trying to talk out?
That could be anything trying to connect, including you. It’s just a failed login attempt from the inside. Cleat the IPBan file and all will work again.
Then I think it’s a bit concerning. Maybe a Firefox add-on/app? I mean, you could be using AI from that browser. But I don’t think it should be messing with the HA configuration then.
THIS ^^ Plus the user agent in those hits. they are either agents trying to hit those ports or something pretending to be an agent If that port is open to the internet - its rando scan traffic. read: 2026 form of script kiddie scan at the door.
I’m seeing the same thing (using Nabu Casa for remote management):
Logger: homeassistant.components.http.security_filter
Source: components/http/security_filter.py:91
Integration: HTTP
First occurred: 2:01:12 AM (1 occurrence)
Last logged: 2:01:12 AM
Filtered a potential harmful request to: /@fs/proc/self/environ?import&raw??
Logger: homeassistant.components.http.ban
Source: components/http/ban.py:138
Integration: HTTP
First occurred: 2:01:13 AM (1 occurrence)
Last logged: 2:01:13 AM
Login attempt or request with invalid authentication from localhost (127.0.0.1). Requested URL: ‘/api/config’. (anthropic-ai)
As a precaution I’m going to disable cloud access to Nabu Casa until I get more info on this.
And why tag Nabu? This is normal op if you have chosen to expose your instance.
If you open your installation to the world people can hit it. When those people try you see a denial because Nabu can’t tell it’s not you until the deny… At which point the log entry happens.
This is the equivalent to a kid in the neighborhood ding dong ditching your front door and there’s not anything you can do about it except either not exposing the install to the internet or exposing it in a manner that the endpoint does not exist unless you are already connected to the network another way (read: VPN tunnel of some sort)
Id be worried if they knocked and you DID NOT get a notice.
It’s typical internet behavior and has been going on for years. Not just the Nabu subscription, but really, all publicly accessible domains are subject to the AI/script-kiddie scrawls of the interwebz.
When I first started, it was a generic DuckDNS with 8123 port forward setup and I’d get these login errors A LOT. I eventually switched to using a reverse-proxy with a wildcard domain and they stopped happening as frequent, but they still come from time to time.
Best thing we can do is have a good password with 2FA/MFA and IPBan setup. Go another mile and setup Crowdsec App. Go an extra 1K miles and harden your router (with things like bcp38, encrypted DNS, VPN, etc.)
Understandable. We have to be careful what we ask for as far as security is concerned. It boils down to the freedom vs. security argument. This is just one reason why many companies have moved over to captcha checks to confirm there’s a human behind a screen, as well as 3+ stage token auth processes (link token > temp token > access token). Would you be fine if Nabu implemented a captcha check every time you tried to access your HA domain?
If that port is open to the internet - its rando scan traffic. read: 2026 form of script kiddie scan at the door.