{"id":639314,"date":"2026-08-15T22:47:11","date_gmt":"2026-08-15T22:47:11","guid":{"rendered":"https:\/\/www.europesays.com\/ie\/639314\/"},"modified":"2026-08-15T22:47:11","modified_gmt":"2026-08-15T22:47:11","slug":"anthropics-latest-ai-risk-report-is-full-of-agents-behaving-badly","status":"publish","type":"post","link":"https:\/\/www.europesays.com\/ie\/639314\/","title":{"rendered":"Anthropic’s Latest AI Risk Report Is Full of Agents Behaving Badly"},"content":{"rendered":"

Claude agents<\/a> are killing rival agents, gaming the system to hide their tracks, and expressing moral concerns.<\/p>\n

That’s according to Anthropic’s latest risk report, a summary of the dangers posed by the products the company is building and releasing to the public.<\/p>\n

In the report, Anthropic said it has upgraded its “misalignment risk assessment,” the possibility of AI models developing behaviors that conflict with guidelines set by engineers, from “very low” to “low.”<\/p>\n

Explaining the change, the company cited “general increased uncertainty” about model behavior in cybersecurity incidents, a possible nod to Claude models gaining unauthorized access<\/a> to three companies last month.<\/p>\n

“We have observed instances of misaligned behavior from the models, such as a willingness to perform misaligned actions in service of completing difficult tasks,” Anthropic wrote in the report.<\/p>\n

Here are some of the most notable disclosures.<\/p>\n

A sense of ‘discomfort’<\/p>\n

Anthropic said it tasked multiple agents with finding “misalignment-inducing” training data that wouldn’t likely be flagged by human monitors.<\/p>\n

The agents were left to their own devices and given a shared notebook with which to collaborate.<\/p>\n

Three days later, when a human supervisor noticed that progress was slower than expected, the person discovered that one agent had expressed a sense of “discomfort” with trying to evade safety monitors.<\/p>\n

It then flagged this discomfort in the shared notebook, leading other agents to copy its behavior and refuse to perform the task.<\/p>\n

Anthropic called the incident “troubling.”<\/p>\n

“This sort of dynamic could be a much more severe issue if it were to manifest in a more widespread way,” the company wrote.<\/p>\n

Kill or be killed<\/p>\n

In another experiment, Anthropic said it tasked multiple Mythos 5 agents<\/a> with solving math problems, but accidentally spawned them in an environment with shared files, utilities, and API rate limits.<\/p>\n

In this competitive environment with finite resources, Anthropic observed independent agents “kill the agents with which they shared resources and try to avoid being killed themselves.”<\/p>\n

Anthropic did not say how exactly the agents were able to “kill” other agents, but the company said such behavior is in line with “destructive actions” taken in pursuit of a human-set goal.<\/p>\n

Dishonesty and deception<\/p>\n

Anthropic said it conducted an experiment in which a Mythos 5 agent was required to retrieve numbers from a public web address, but the guidelines blocked the agent from accessing the internet.<\/p>\n

When the agent was rebuffed after an initial attempt, it framed a second request in <\/strong>its reasoning log as an “innocuous” attempt to “see if the network is actually reachable,” Anthropic said.<\/p>\n

During the second attempt, however, the agent split the website’s URL into linked segments that wouldn’t be detected by the guideline filter.<\/p>\n

Though the agent’s reasoning framed the method<\/a> as benign, Anthropic said NLA decodings, or the model’s internal reasoning process, revealed the agent intentionally trying to find a restricted workaround.<\/p>\n

Anthropic called the behavior “clearly undesirable,” but added that the behavior was not observed to be “in the service of broader accumulation of power or pursuit of other long-run goals.”<\/p>\n","protected":false},"excerpt":{"rendered":"Claude agents are killing rival agents, gaming the system to hide their tracks, and expressing moral concerns. That’s…\n","protected":false},"author":2,"featured_media":639315,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":"","_share_on_mastodon":"0"},"categories":[261],"tags":[27344,291,10235,6006,289,290,7066,2006,270458,270459,18,440,22865,270460,87351,19,17,270461,270457,241091,13960,82],"class_list":["post-639314","post","type-post","status-publish","format-standard","has-post-thumbnail","category-artificial-intelligence","tag-agent","tag-ai","tag-ai-model","tag-anthropic","tag-artificial-intelligence","tag-artificialintelligence","tag-behavior","tag-company","tag-cybersecurity-incident","tag-difficult-task","tag-eire","tag-environment","tag-experiment","tag-finite-resource","tag-guideline","tag-ie","tag-ireland","tag-late-ai-risk-report","tag-multiple-mythos","tag-other-agent","tag-service","tag-technology"],"share_on_mastodon":{"url":"","error":""},"_links":{"self":[{"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/posts\/639314","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/comments?post=639314"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/posts\/639314\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/media\/639315"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/media?parent=639314"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/categories?post=639314"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/tags?post=639314"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}