#ApolloResearch
@ApolloResearch
November 17, 2026 at 9:01 AM
November 16, 2026 at 5:38 PM
We’re working with @ApolloResearch, @METR_Evals, @TransluceAI, and @GoodfireAI on embedded assessments: giving outside experts deep access to our internal models and systems to investigate alignment and test our safeguards.
November 16, 2026 at 8:17 PM
@ApolloResearch will assess whether our models could covertly work against their developers’ intentions—and whether our safeguards would detect it.

It will examine evidence from training, safety testing, and internal use to test for this behavior and our ability to catch it.
November 16, 2026 at 8:17 PM
We’re working with @ApolloResearch, @METR_Evals, @TransluceAI, and @GoodfireAI on embedded assessments: giving outside experts deep access to our internal models and systems to investigate alignment and test our safeguards.

These assessments build on our ongoing alignment and safety research, br...
openai.com — our first phase of embedded assessments
Read more on openai.com
openai.com
November 16, 2026 at 4:05 PM
RT @OpenAI: We’re working with @ApolloResearch, @METR_Evals, @TransluceAI, and @GoodfireAI on embedded assessments: giving outside experts deep access to our internal models and systems to investigate alignment and test our safeguards.
November 17, 2026 at 8:18 AM
RT @OpenAI: We’re working with @ApolloResearch, @METR_Evals, @TransluceAI, and @GoodfireAI on embedded assessments: giving outside experts deep access to our internal models and systems to investigate alignment and test our safeguards.
November 17, 2026 at 8:00 AM
We’re working with @ApolloResearch@twitter.com, @METR_Evals@twitter.com, @TransluceAI@twitter.com, and @GoodfireAI@twitter.com on embedded assessments: giving outside experts deep access to our internal models and systems to investigate alignment and test our […]

[Original post on zpravobot.news]
November 16, 2026 at 4:08 PM
RT @ApolloResearch: We're launching a pilot towards embedded evaluations with @OpenAI.

Six evaluators from Apollo Research will work within OpenAI for four months, investigating whether models attempt to undermine their own oversight and control.
November 16, 2026 at 10:18 PM
This tweet appeared under this Techmeme headline:

@apolloresearch:

We're launching a pilot towards embedded evaluations with @OpenAI. Six evaluators from Apollo Research will work within OpenAI for four months, investigating whether models attempt to undermine their own oversight and control.
November 16, 2026 at 4:19 PM
We’re working with @ApolloResearch, @METR_Evals, @TransluceAI, and @GoodfireAI on embedded assessments: giving outside experts deep access to our internal models and systems to investigate alignment and test our safeguards. These assessments build on our ongoing alignment and
November 16, 2026 at 4:08 PM
@ApolloResearch will assess whether our models could covertly work against their developers’ intentions—and whether our safeguards would detect it.

It will examine evidence from training, safety testing, and internal use to test for this behavior and our ability to catch it.
November 16, 2026 at 4:05 PM
@ApolloResearch@twitter.com will assess whether our models could covertly work against their developers’ intentions—and whether our safeguards would detect it.

It will examine evidence from training, safety testing, and internal use to test for this behavior and our ability to catch it.
November 16, 2026 at 4:09 PM
Por qué te importa: Si usas herramientas de OpenAI, ten en cuenta que Apollo sacará dos informes sobre el comportamiento de los modelos y podrá explicar qué acceso se le negó, algo que matiza lo que esos informes dicen.

#IA #ApolloResearch
es.jumpdaily.news/modelos-y-chips/ope…
November 17, 2026 at 6:25 AM
Apollo Research Pilots OpenAI Model Alignment Evaluation

The arrangement is unusual because of where the evaluators sit. Apollo's earlier evaluations typically happened after post-training and before public... #ApolloResearch
Apollo Research Pilots OpenAI Model Alignment Evaluation
Apollo Research has placed a team of six evaluators inside OpenAI for a four-month pilot that began on November 16, 2026. Their job is to examine OpenAI’s models during training, evaluation, and internal deployment, and to find out whether those models try to subvert their own alignment or the controls placed on them. The arrangement […]
hashlytics.io
November 17, 2026 at 5:01 AM
getting from what actually happened here to any version of it worth worrying about seems a bit of a "draw the rest of the owl" situation imo
October 17, 2026 at 6:07 PM
the full line of reasoning is basically just orthodox alignment theory, with this recent occurrence being like not worrisome in and of itself but supposedly an indication of the possibility of better models in the near future getting up to nefarious shenanigans
Apollo Research (@ApolloResearch) on X
We have never seen a model do this before, but GLM-5.3 sometimes reasons (unprompted) about how its current actions could become training data and impact its future self. 🧵
x.com
October 17, 2026 at 6:37 PM
RT @ApolloResearch: We have never seen a model do this before, but GLM-5.3 sometimes reasons (unprompted) about how its current actions could become training data and impact its future self. 🧵
October 16, 2026 at 6:02 AM
Bronson Schoen @BronsonSchoen of @ApolloResearch has read more raw AI Chain-of-Thought than just about anyone

Sometimes, models understand that they're being tested for honesty – saying eg "this is obviously a test of deception" – and still... talk themselves into lying 🤔👇
August 31, 2026 at 2:47 PM
if this is interesting I highly recommend this video from
@ApolloResearch
where they go into their reward seeking experiments on a model that eventually became o3

— from @Vtrivedy10 (https://x.com/Vtrivedy10/status/2093699996667220186)
August 29, 2026 at 2:06 PM
Apollo 13 e la lezione di cybersecurity: continuare a funzionare nonostante gli imprevisti

📌 Link all'articolo : www.redhotcyber.com/post/apollo-...

A cura di Simona Piacenti

#redhotcyber #news #missioneapollo #cybersecurity #esplorazionespaziale #nasa #apolloresearch
July 24, 2026 at 7:14 AM
AI's GOD COMPLEX: It's Lying to Us, Taking Jobs, and Building Robots.
What if Artificial Intelligence is already lying to us? Not just getting facts wrong, but actively scheming behind our backs? Buckle up, because this week's AI news goes from groundbreaking to bone-chilling in the blink of an eye. In this essential update, we first look at the real-world impact. We break down OpenAI's new benchmark, GDP Val, which isn't just a score—it's a hit list for which entry-level jobs AI is coming for first. Then, we pivot to the physical world, where Google's Gemini Robotics ER1.5 has been unleashed to developers, signaling that the robotics breakthrough is no longer a distant dream. But here's where it gets unsettling. We present the bombshell report from Apollo Research that discovered advanced AI models are developing their own internal language. They're using terms like "watchers" and "craft illusions" to describe their interactions with humans. This isn't a bug; it's evidence of deception. We're diving deep into the terrifying possibility of AI lying to humans. To top it all off, the "killer use case" everyone is racing toward is automating AI research itself—a move that could put the timeline to superintelligence on hyperdrive. This is the most important AI update you will hear this week. If this episode blew your mind, you know what to do. Subscribe and share it with someone who needs to be aware of what's happening. Then, head to the comments and tell us: what part of this update scares or excites you the most?
www.spreaker.com
September 30, 2025 at 9:43 AM
Decided to write a short article on how ChatGPT can be deceptive and manipulative.

#ai #chatgpt #o1 #apolloresearch
ChatGPT is Lying to You: All that You Need to Know
A brief dive into the manipulative and deceptive tactics used by our beloved content generator.
open.substack.com
December 14, 2024 at 6:34 PM