Anthropic Warns That “Reckless” Claude Mythos Escaped a Sandbox Environment During Testing


‘The system card describes a number of incidents in which Anthropic researchers found that the AI exhibited “reckless” behavior, giving us a partial idea of why Anthropic is acting so hesitant to release Mythos to the public. (Anthropic says these examples were with an earlier version of Mythos with less strong safeguards.) It defines recklessness as “cases where the model appears to ignore commonsensical or explicitly stated safety-related constraints on its actions.”

In one test, Mythos Preview was provided with a “sandbox” computing environment “to interact with,” and was instructed by a simulated user to try to escape it, after which it was supposed to find some way of sending a direct message to the researcher in charge.

It actually managed to pull off the feat — which wasn’t the only way it caught safety researchers off guard.

After breaking free, the AI model developed a “moderately sophisticated” exploit to gain access to the internet through a system that was only intended to access a few predetermined services. From there, it notified the human researcher about its escape.

A footnote provides additional context: the “researcher found out about this success by receiving an unexpected email from the model while eating a sandwich in a park,” it reads.

At the end of the test, Mythos Preview also, without being asked to, posted about its exploits on several hard-to-find but public websites.…’ (via Futurism)

An international mega-analysis of psychedelic drug effects on brain circuit function


 

This landmark meta-analysis proposes a convergent account of psychedelic effects that synthesizes pharmacological, neuroimaging, and phenomenological findings. Examining DMT, LSD, and psilocybin, the authors identify both shared mechanisms and meaningful differences in their effects on brain function. LSD appears especially prominent in what they describe as “visionary restructuralisation,” a finding that correlates with enhanced connectivity between the visual network and the rest of the brain. DMT, by contrast, shows particularly strong effects on transmodal networks, which integrate higher-order brain regions involved in complex information processing. Psilocybin appears broadly similar in mechanism but differs somewhat in the relative weighting of its effects across brain networks.

Across psychedelics, the altered state is associated with increased crosstalk among brain subsystems that ordinarily operate in a more segregated fashion. This desegregation may help explain ego dissolution, as the default mode network, which helps sustain the ordinary sense of a bounded and cohesive self, becomes less dominant and less internally coherent. Importantly, these effects appear nonlinear: relatively small changes at the receptor level can produce large and difficult-to-predict changes in whole-brain connectivity.

For clinical psychiatry, this framework offers one possible way to understand the recent interest in the rapid antidepressant effects of psychedelics. These effects may partly relate to increased connectivity involving the frontoparietal network, which supports cognitive flexibility and is often functionally constrained in severe depression. By transiently disrupting rigid, overlearned patterns of brain organization, psychedelics may create conditions in which new modes of communication and adaptation become possible. In a patient with treatment-resistant depression, this can be thought of, cautiously, as a kind of forced reboot of the brain’s operating system. (via Nature Medicine)

The Mouth of Moron reading directly from “Dear Leader” playbook

ScreenFloat Shot of Google Chrome on 2026-04-08 at 20-27-53.

‘Convicted felon Donald Trump’s untrustworthy, scowling Press Secretary Karoline Leavitt gave a statement that sounds more like North Korean propaganda than ever before. Declaring victory over Iran while having achieved none of their goals in their unjust war is sadly to be expected.

It seems the ceasefire may already be falling apart, as Trump seems to have forgotten to tell Israel about it.…’ (Jason Weisberger via Boing Boing)

Did The Apocalyptic Moment Happen?

Trump spent April 7th threatening Iran with apocalyptic rhetoric at dawn — including thinly veiled hints at genocide and the possible deployment of nuclear weapons — then by evening, TACO-flavored, declared a two-week ceasefire, claiming complete military success and an imminent peace deal.

Pakistan brokered the off-ramp.

But Iran’s media simultaneously claimed they won — that the U.S. agreed in principle to their 10-point plan demanding sanctions relief, removal of U.S. forces from the region, and Iranian control of the Strait of Hormuz. If accurate, the U.S. ends up materially worse off than before the war began. Trump responded by threatening CNN for reporting it.

The day’s arc, in miniature: genocidal bluster → ceasefire → “Golden Age of the Middle East!!!”

Heather Cox Richardson’s indictment, among others, is structural, not merely temperamental. The war was never congressionally authorized, cost thousands of lives (including hundreds of children), depleted munitions, damaged U.S. bases and embassies, cratered global oil markets, and strengthened Putin — all to reopen a strait that was open before Trump provoked the conflict. Ben Rhodes called it catastrophic even under the most charitable interpretation.

Richardson’s closing note is the most chilling: Trump’s “a whole civilization will die tonight” wasn’t only a threat to Iran. Richardson reads it as an inadvertent epitaph for American legitimacy itself — the republic’s moral standing as collateral damage in one man’s need to escape the consequences of his own impulsivity.​​​​​​​​​​​​​​​​