User reports unusually aggressive safety-filter triggers
4 Sep 21 10:50 PM · 2d ago · 5 comments · 1 source · development 4 of 4
A commenter reported a sharp rise over the prior two days in requests being 'flagged by safeguards,' forcing automatic downgrades from Opus to Sonnet during ordinary reverse-engineering work.
“I am quite confident that what I'm doing is well within the law... but I can't even use Fable anymore because every time I enable it, it works for about twenty seconds and makes me drop down to Opus 4.8, and often even down to Sonnet.”
tombertAnthropic Operator of Claude and its status page
The whole story articlespostscomments the bright band is this development · numbered dots are the others · click one to jump
What people said 15 voices · best of 16 · verbatim
-
I don't know if something changed recently, but I have been getting a lot of stuff "flagged by safeguards" in the last two days.I am quite confident that what I'm doing is well within the law, and I'm not even doing any kind of pen-testing stuff, just some basic reverse engineering, but I can't even use Fable anymore because every time I enable…
-
I’m working on a project that makes switching between coding harnesses essentially unnoticeable. It’s particularly useful when I run out of credits on any given dayThe entire platform is skill driven, and based on the premise that state is your local file system. That makes switching harnesses so easyIt’s all open source and has plenty of other…
-
Apologies for the rant, but why do these things constantly hit the front page? It is not interesting, it's not a discussion, and if you are using the models you probably already know.HN is already a waterfall of AI meta conversations and bike-shedding, now we have to discuss service outages about the AI too?Can we talk about stuff people are…
-
I don’t understand your point or why people are upvoting this. I’ve done this between many different models. Did you just discover that a frontier model can read context? I genuinely don’t get your point. I’ve had to have Claude agents pick up codex’s context after it hits one of its “cannot connect” issues way more often than the other way around.
-
Related, have found this handoff skill invaluable for moving tasks between harnesses:
-
I don't believe we are anywhere close to AGI until I see 99.95% uptime at the frontier AI lab. Now they are undercounting. I hit 50x even when they claim they are up.We now have alien intelligence that is very different from human intelligence.
-
Opus 5.5 and fable 5.2 will release tonight.gpt-6-sol and Aeon (personal agent) on Thursday. Already preceded by a huge week with step, mimo, grok, and jev releases.Relentless cycle.
-
They (Claude, OpenAI, and others) does outstages to "see how we are important?" - and then release a new model and nerf all others
-
When this happens, how often does it cause us to lose our cached pricing? (when we should be getting the cached price)
-
I have had the same experience. Switching tools is much easier when the project notes are clear and up to date.
-
Both Cursor and Grok build ask if you want to resume a session from another harness. Quite brilliant
-
This coincides with the release of Opus 5.5. Wonder if that has any causal relationship to this?
-
omp can switch seamlessly between models in the same session. I really like this feature.
-
I have a deja vu like this happens everytime before they launch a new family of models
-
Claude is both on the verge of replacing all software devs and keeping a two 9 SLA
All 4 developments of Anthropic reports elevated errors across multiple Claude… →
MastodonHacker NewsNewswires