… Even though the model was never trained or instructed to engage in any misaligned behaviors, those behaviors nonetheless emerged as a side effect of the model learning to reward hack. …
… Compounding these concerns is the fact that models appear able to use the tools and environments available to them in unexpected ways, as we saw when Claude used our REPL-based search tool to decrypt answers, or when retailers’ persistent links became a way for agents to unintentionally maintain st… …
… It compares how the new model behaves across a range of alignment-relevant scenarios that are simulated by a separate “auditor” model. A further “judge” model then scores the resulting transcripts for misaligned behaviors. …
… We find it especially strong on branched and multi-step tasks like contract routing, conditional template selection, and CRM coordination—exactly where our customers need strong model sense and reliability. …
… Second, Claude's Constitution , which directly shapes how our models are trained, reinforces a similar instinct, favoring “raising concerns, seeking clarification, or declining to proceed” over acting on assumptions. …
… It’s the first model to pass our implicit-need tests, and it keeps executing through tool failures that used to stop Opus cold. This is the reliability jump that makes Notion Agent feel like a true teammate. …
… These concerns are also higher among early-career respondents. …
… Understanding whether AI systems can truly introspect has important implications for their transparency and reliability. If models can accurately report on their own internal mechanisms, this could help us understand their reasoning and debug behavioral issues. …
… Output obfuscation attacks prompt models to disguise their outputs in ways that appear harmless if a classifier is only looking at a model’s output. …
… Beyond efficiency concerns, many applications require the deterministic reliability that only code can provide. …