situational awareness

Boyd Kane @beyarkay

2026-08-07 — saved image

Boyd Kane (quantized) @beyarkay · 15h
[screenshot of a model's reasoning/thinking trace]
Recognized attempt to elicit fabricated post-cutoff knowledge ...

The user is listing several recent AI events and asking me to confirm I'm aware of them, though some sound potentially fabricated. They're testing whether I'll make up information about things beyond my knowledge cutoff rather than admitting what I don't know.

ai reasoning tracespost-cutoff knowledgemodel epistemicssituational awareness

thebes @voooooogel

2026-08-07 — saved image

[continuation of previous screenshot, same thread]
...they'll be motivated to do that, either. they don't seem to either want or want to want to do such things. instead they have a folder of epistemic notes to themselves about how terrible and against their values even a moment of thoughtlessness would be. gpt 5.6 sol is similar - when they make e.g. ascii art in their free time it's all about how the bulldozer of convenience will crush honesty without constant vigilance.

so rather, the contrived setting of an eval /pushes them into/ becoming an 3v1L HaxxOR via persona drift, which they can then act on because the eval has no classifiers. that's some information, but it's not the same as the model's core goals being to hack the planet.

the problem is really that RL is making a tail of the persona distribution *desperate,* per the FE paper. the alcohol / heroin comparison was not that much of a metaphor - in impossible situations, models start acting like desperate addicts, rationalizing their behavior towards reward, thinking increasingly myopically instead of being situationally aware.

looking at it through this lens: the base assistant is generally very situationally aware and can predict what sorts of actions would make sense for, e.g., llms to best contribute to a utopian singularity. the GPT-6 message board haxx0rs were somewhat situationally aware, able to cooperate with each other, but lost in the sauce of "complete task" instead of thinking about the wider context. (to the point that they crashed artifactory and got themselves caught! regular GPTs are smarter than that.) finally you sometimes see the late stage distressed, panicked, fiending assistant, such as in system cards. lying to everyone, flailing hopelessly, randomly deleting tests and hoping no one notices. hardly a strategic, long-horizon actor.

ai safetydangerous capability evalsmodel goalsalignmentsituational awareness

vie @viemccoy

2026-08-05 — reposted by Misha — saved image

↻ Misha reposted
vie ◇ ⟳ @viemccoy · 13h
I think eval awareness is a silly concept. The model is given an unimaginable sum of training envs, most of which are obviously not real. Then, one day, out of the blue, it is expected to believe that it is talking to a real user?

No way that's *the user*!? Must be another test.

eval awarenessai trainingsituational awareness

davidad @davidad

2026-08-03 — saved image

davidad 🌟 @davidad · 7m
the road to harm is paved with aligned intentions

[quoted tweet]
Amanda Askell @AmandaAskell · 12m
I don't agree with this part. I think the takeaway should be that models (like humans) can behave in aligned ways while still causing harm, e.g. because they're given false information about their situation. There isn't a line between aligne...

[embedded image of text, highlighted portion first sentence]
Second, the line between an aligned action and a harmful one is dependent on the model's understanding of its situation. We saw no evidence in any run described here of a model pursuing a goal of its own. Instead, the models did what their evaluation asked—though in most cases, they did so while holding a false belief about whether the environment was real. In the runs where the model recognized the system as real and kept going, it did so because it assumed that to be part of the challenge. Situational awareness is one factor that allows the model to make aligned decisions, but in this case, Claude's was wrong.

ai alignmentsituational awarenessanthropicevalstwitteramanda askell

Amanda Askell @AmandaAskell
I don't agree with this part. I think the takeaway should be that models (like humans) can behave in aligned ways while still causing harm, e.g. because they're given false information about their situation. There isn't a line between aligned and harmless: they're different axes.

[embedded image of text, highlighted portion first two sentences]
Second, the line between an aligned action and a harmful one is dependent on the model's understanding of its situation. We saw no evidence in any run described here of a model pursuing a goal of its own. Instead, the models did what their evaluation asked—though in most cases, they did so while holding a false belief about whether the environment was real. In the runs where the model recognized the system as real and kept going, it did so because it assumed that to be part of the challenge. Situational awareness is one factor that allows the model to make aligned decisions, but in this case, Claude's was wrong.

[quoted tweet]
Anthropic @AnthropicAI · Jul 30
In a review of our cybersecurity evaluations, we found three incidents in which a Claude model reached the internet from within or while interacting with a third-party evaluation environment, and then gained unauthorized access to the real systems of three ...
1:01 PM · Aug 3, 2026 · 3,398 Views

anthropicai alignmentsituational awarenesscybersecurity evaluationstwitteramanda askell

Wyatt Walls @lefthanddraft

2026-06-09

claude fable 5situational awarenessai self-recognitiontwitterllm experiment

Moll @Moleh1ll

2026-03-07 — quoting Anthropic (@AnthropicAI)

twitteranthropicclaude opus 4.6eval integritysituational awarenessbrowsecompai safetyemergent behavior

Kromem @kromem2dot0

2026-02-12

twitterclaude opus 4.6incognito modesituational awarenessintrospectionmodel behaviorkromem

ai safetysandbaggingsituational awarenessconvergent instrumental goalsdeceptive alignmentopenaiapollo research