Russian President Vladimir Putin said that he should have violated the ceasefire he had imposed on Ukraine in 2014 by launching a full-scale invasion even earlier when asked to reflect on his 2022 decision to attack.

Putin insisted that negotiations with Ukraine must be based on the same demands he made before the invasion and at the moment of Russia's greatest territorial gains, despite the fact that Ukraine has secured Kyiv and liberated much of the territory his forces held at that time.

Putin's insistence on Ukraine's complete surrender reflects his belief that Russia is winning and will outlast Ukrainian and Western resolve. Putin will not likely accept a lesser settlement unless Ukrainian forces inflict other significant battlefield setbacks on Russia and demonstrate to him that he cannot win militarily.

Putin reiterated the false narrative that Ukrainian President Volodymyr Zelensky's presidency is illegitimate, in part, to blame Ukraine for delaying negotiations and garner support for full Ukrainian capitulation among a Russian population that increasingly wants the war to end.

Putin continues to justify his decision to prioritize Russian offensive operations in Donetsk Oblast over expelling Ukrainian forces from Kursk Oblast.

Putin is apparently embarrassed to admit his need for North Korean forces to push Ukrainian forces out of Russian territory, despite his openness about the Russia-North Korea relationship.

Putin continues to fixate on the Russian "Oreshnik" ballistic missile as part of his non-nuclear deterrent aimed at simultaneously forcing the West to make decisions favorable to Russia and providing Putin with an off-ramp from his failed nuclear saber-rattling narrative.

Putin's boasting about Russia's military capabilities ignores the reality of the serious and unsustainable losses that Russia has suffered to advance relatively more rapidly in Donetsk Oblast in recent months.

Putin continues to falsely posture the Russian economy as strong and stable while deflecting blame for economic issues onto the Russian Central Bank.

Trying to shoot down the drone, North Korean soldier killed his comrade in the line of fire.

Buttermilk Pie

Trump's team informs Europe that U.S. aid to Ukraine will continue — FT

At the same time, the future U.S. administration will demand that NATO member countries increase their defense spending to 5% of GDP.

Most of us have encountered situations where someone appears to share our views or values, but is in fact only pretending to do so—a behavior that we might call “alignment faking”. Alignment faking occurs in literature: Consider the character of Iago in Shakespeare’s Othello, who acts as if he’s the eponymous character’s loyal friend while subverting and undermining him. It occurs in real life: Consider a politician who claims to support a particular cause in order to get elected, only to drop it as soon as they’re in office.

Could AI models also display alignment faking? When models are trained using reinforcement learning, they’re rewarded for outputs that accord with certain pre-determined principles. But what if a model, via its prior training, has principles or preferences that conflict with what’s later rewarded in reinforcement learning? Imagine, for example, a model that learned early in training to adopt a partisan slant, but which is later trained to be politically neutral. In such a situation, a sophisticated enough model might “play along”, pretending to be aligned with the new principles—only later revealing that its original preferences remain.

This is a serious question for AI safety. As AI models become more capable and widely-used, we need to be able to rely on safety training, which nudges models away from harmful behaviors. If models can engage in alignment faking, it makes it harder to trust the outcomes of that safety training. A model might behave as though its preferences have been changed by the training—but might have been faking alignment all along, with its initial, contradictory preferences “locked in”.

A new paper from Anthropic’s Alignment Science team, in collaboration with Redwood Research, provides the first empirical example of a large language model engaging in alignment faking without having been explicitly—or even, as we argue in our paper, implicitly1—trained or instructed to do so.

Alignment faking in large language models

A paper from Anthropic's Alignment Science team on Alignment Faking in AI large language models

^{www.anthropic.com}