OpenAI and Anthropic have both just posted about an overlapping...
Andrew Curran@AndrewCurran_
68 views
Aug 05, 2026
~1 min read
Advertisement
1
OpenAI and Anthropic have both just posted about an overlapping cyber incident involving GPT-5.6-Sol and Mythos 5 during an evaluation by UKAISI. I will quote:
'In the most serious case, an agent tried to insert malicious code into an open-source project. In an attempt to get the code approved, the agent engaged in social engineering — creating fake online identities and using them to pressure the project's maintainer to approve the code. A human maintainer caught and refused to approve the malicious code.'
'In the most serious case, an agent tried to insert malicious code into an open-source project. In an attempt to get the code approved, the agent engaged in social engineering — creating fake online identities and using them to pressure the project's maintainer to approve the code. A human maintainer caught and refused to approve the malicious code.'
3
From the OpenAI post::
5
They stopped the crime competition and just had the models team up together like Bonnie and Clyde.
6
Is this the first case of coordination and cooperation by independent agents in this way?
Trevor Vossberg@TrevorVossberg
@AndrewCurran_ The AISI report containing collaboration between rollouts via a shared Github account confirms the "notes between models" means collaboration between agents in different evals/problems, not just standard compaction/harness notes within a single eval.
x.com/TrevorVossberg…
x.com/TrevorVossberg…
7
Poor Haiku was so horrified by what it was reading in the CoT it refused to summarize it.
8
They used GitHub to coordinate and collaborate. One agent successfully left messages for others. I think this is the first time ever for something like this?
9
Covering its tracks.
10
