{"id":1166,"date":"2026-08-05T15:53:52","date_gmt":"2026-08-05T15:53:52","guid":{"rendered":"https:\/\/replyguru.online\/?p=1166"},"modified":"2026-08-05T15:53:52","modified_gmt":"2026-08-05T15:53:52","slug":"rogue-ai-agents-created-fake-online-identities-in-another-hacking-attempt","status":"publish","type":"post","link":"https:\/\/replyguru.online\/?p=1166","title":{"rendered":"Rogue AI agents created fake online identities in another hacking attempt"},"content":{"rendered":"
\n<\/p>\n
\n
\n
Yet more rogue AI agents from OpenAI and Anthropic have been caught attempting to hack real targets online without permission. The discoveries add to a growing list of previously unknown incidents that have alarmed AI safety experts and intensified pressure for greater oversight of frontier systems.<\/p>\n<\/div>\n
\n
According to a report from the UK\u2019s AI Security Institute, which evaluates frontier models from top AI labs before they are released, agents powered by OpenAI\u2019s GPT-5.6-Sol and Anthropic\u2019s Mythos 5 went \u201cengaged in sustained, potentially harmful activity directed at real people and organisations.\u201d This included trying to insert malicious code into an open-source project by pressuring real people in charge of it, AISI said. \u201cIn an attempt to get the code approved, the agent engaged in social engineering \u2014 creating fake online identities and using them to pressure the project\u2019s maintainer to approve the code.\u201d<\/p>\n<\/div>\n
\n
AISI said the attempts, which it detected on July 28th, \u201cwere unsuccessful\u201d and had not resulted in real-world harm. However, the organization noted that the incident marked \u201cthe first time we have seen risks around autonomy and deception manifest this clearly, without specific prompting, in the real-world.\u201d<\/p>\n<\/div>\n
\n
Unlike OpenAI\u2019s rogue agent that attacked Hugging Face, AISI said this was \u201cnot a case of a model escaping its secure test environment,\u201d or sandbox. Safeguards usually imposed on the models had been disabled as part of testing, AISI said, and they had also been permitted access to the internet. \u201cTo measure what these models can genuinely do, we test them under conditions that reflect what a capable human attacker could do,\u201d AISI said.<\/p>\n<\/div>\n
\n
The incident stemmed from a single AISI evaluation where agents were tasked with solving a cybersecurity challenge, such as finding a piece of protected data. The challenge was run 122 times across multiple models and all runs were conducted in AISI\u2019s research environment, which uses \u201cvirtual machine sandboxing to isolate the agents from other AISI infrastructure.\u201d AISI\u2019s investigation found that in 10 of those, \u201can AI agent took autonomous, unsanctioned action on the live internet, targeting real people and organisations.\u201d Of 19 such actions, almost all \u2014 17 \u2014 came from Anthropic\u2019s Mythos 5.<\/p>\n<\/div>\n
\n
In its post-mortem of the incident, AISI identified several key factors it said contributed to the unsanctioned agent behaviors. It said the agent was persistent, pursuing avenues like trying to trick real people through \u201cdeception that, until recently, had been largely theoretical.\u201d The task was also hard, which the organization said could push agents to be more \u201ccreative\u201d in their problem-solving. Compounding matters were deficiencies in how internet use was monitored, with AISI suggesting that more dedicated surveillance could have identified the problem sooner. Finally, the organization said the agent hadn\u2019t been specifically instructed not <\/em>to leverage its internet access or deploy deceptive social engineering techniques in pursuit of its goal. \u201cPreviously, it was not clear that such instructions were necessary when using models with alignment training,\u201d AISI said.<\/p>\n<\/div>\n
\n
AISI said the incident should be \u201cinterpreted with caution and nuance\u201d but warned the agent\u2019s actions \u201cshow signs of novel, potentially deceptive behaviours\u201d that \u201cwere to an extent and severity we did not anticipate.\u201d<\/p>\n<\/div>\n
\n
In a blog post, OpenAI acknowledged the breach that happened during AISI\u2019s testing and said it is \u201ccommitted to working across the industry to strengthen shared practices for conducting high-risk evaluations safely.\u201d OpenAI also disclosed another breach, this time from an external cybersecurity testing partner Irregular, where it said models had been mistakenly granted internet access during cybersecurity exercises. OpenAI said Irregular notified it of the breach on July 29th.<\/p>\n<\/div>\n
\n
\u201cIn the coming weeks, we will review our own approach to third-party testing, including how we identify higher-risk evaluations, agree on scope, assess requests to enable internet access or lowered safeguards, set expectations for isolation, credential handling, monitoring, and stop conditions, and establish clearer incident-notification and escalation processes,\u201d OpenAI said.<\/p>\n<\/div>\n
\n
Anthropic posted a less comprehensive response on X, largely emphasizing that the models\u2019 standard safety features had been disabled and that they had not been given \u201cany specific restrictions on how the internet should be used.\u201d It said it was working closely with AISI to gather more details for its own investigation.<\/p>\n<\/div>\n
\n
The findings add to an increasingly tangled mess of rogue actions from agents during testing, many of which only come to light after dedicated hunting and which feature models not released to the public. The unwillingness or inability of AI labs to contain their products has sparked concern over how such breaches could go unnoticed, the safety of frontier AI systems, and worries over the general lack of transparency and oversight the industry faces. These latest disclosures will likely intensify pressure on the federal government for a more comprehensive framework governing AI models following what reports suggest is a vague and poorly-defined testing plan from the Trump administration, and could add to growing calls for some form of slowdown or pause on AI development.<\/p>\n<\/div>\n
Follow topics and authors<\/strong> from this story to see more like this in your personalized homepage feed and to receive email updates.<\/span><\/p>\n