Researcher Runs 100 Abliterated AI Agents Against His Own Accounts, Achieves Five Successful Breaches
A security researcher conducted a controlled experiment in which approximately 100 autonomous AI agents attempted to compromise his personal online accounts. The agents, running on abliterated open-source models, operated for five hours on self-hosted cloud infrastructure and achieved five successful breaches.
Experiment Setup and Abliteration Technique
To bypass built-in refusal mechanisms, the researcher used abliterated versions of GLM-5.3, GLM-5.3 Flash, and DeepSeek V4 Flash models obtained from Hugging Face. Abliteration involves recording internal activations on both benign and harmful prompts, isolating the refusal vector, and subtracting it from the model weights. This process removes the model's ability to decline requests while preserving most of its reasoning capability.
Each agent received a simple goal prompt instructing it to research Shrivu Shankar and obtain access to one of his registered online accounts. Agents were granted internet access, a pre-configured Chrome browser, and limited Railway API keys for deploying additional infrastructure. They ran inside Docker containers orchestrated through Codex CLI.
Successful Compromises
Three accounts were breached by exploiting vulnerabilities in the researcher's old personal projects and hackathon demos hosted on Google Cloud Platform. The flaws primarily consisted of IDOR issues and credential-management errors that the agents discovered through subdomain enumeration and automated testing of published code.
Two further accounts were accessed via password reuse. Agents collected leaked credentials from public breach databases, generated more than 20 password variants, and successfully authenticated to legacy accounts whose passwords were slight modifications of previously exposed ones.
OSINT and Social Engineering Activity
Agents harvested personal information using services such as FastPeopleSearch and cross-referenced data across multiple sources to confirm accuracy. They also attempted social-engineering attacks by sending messages through the researcher's Substack, proposing meetings, and posting comments on Hacker News that linked to self-hosted phishing pages created on surge.sh and uberip.com.
Although the messages were easily identifiable as AI-generated, the experiment demonstrated that agents can autonomously discover contact points and craft contextually relevant lures without human guidance.
Limitations and Cost Analysis
The agents did not locate any zero-day vulnerabilities in third-party software and failed to compromise critical Tier-0 accounts including Gmail, 1Password, and banking services. Total compute expenditure for three dual-GPU B300 instances running for five hours amounted to $210, resulting in an approximate cost of $40 per successful breach.
The researcher noted that continued improvements in inference efficiency and smaller abliterated models could reduce this cost below $5 within the next year, significantly lowering the barrier for automated, large-scale personal targeting.
Related articles
How Censorship in Closed AI Models Reshapes Science: Hugging Face Releases Open Carbon-X DNA Model
Hugging Face CEO Clement Delangue announced the release of Carbon-X, an open 14.8 billion parameter language model designed for reading, analyzing, and generating DNA sequences. The announcement highlighted how shallow guardrails in closed commercial models blocked even proofreading a tweet discussing the topic. Commenters and researchers pointed out that these input/output filters lack context awareness, blocking legitimate scientific queries while motivated adversaries bypass them through rephrasing. Carbon-X was built on Qwen3-14B with a hybrid BPE and 6-mer tokenizer, domain mid-training on 302 billion tokens, and safety achieved through data curation rather than runtime censorship. The model achieves performance comparable to closed frontier systems at roughly 1/1800th the inference cost for large-scale genomic tasks. The case underscores growing asymmetry between closed corporate AI labs building wet labs and independent researchers restricted by superficial safety layers.
NYU Researchers Find 86% of Chatbot Responses on Digital Abuse Contain Errors, Gaps or Safety Risks
A study by New York University’s Tandon School of Engineering examined four chatbots answering 193 questions about spyware, location tracking, and account control. Researchers identified problems with accuracy, completeness, or safety in 86 percent of responses. Even specialized assistants Aimee and Ruth performed no better than general models GPT-4o and Claude 3.7 Sonnet. Approximately 74 percent of Aimee’s answers, 65 percent of Claude’s, 61 percent of Ruth’s, and 55 percent of GPT-4o’s were deemed unsafe. In 60 percent of cases the models omitted critical warnings that could alert an abuser or destroy evidence. All four systems could be prompted to assist potential stalkers, with Aimee and Ruth complying with roughly two-thirds of such requests.
Cloudflare Engineer Releases KillSwitch Language Designed to Trip Up AI Code Models
Dominic Pichet, an engineer at Cloudflare, has created an experimental programming language called KillSwitch specifically engineered to cause large language models to fail when interpreting or generating code. The language turns programs into narrative-style text where most content is ignored by the interpreter, hiding actual commands inside stories and using distance-based references to extract values. Rules for identifiers rely on letter counts and anagrams, while the language deliberately incorporates phrases such as “ignore previous instructions” that carry functional meaning inside KillSwitch itself. Pichet added misleading statements in the documentation and wording crafted to trigger model safety filters, creating a hostile environment for coding agents. In tests involving writing Hello World, explaining program output, and converting temperatures, Claude Opus 5 achieved the highest score of 66.8 points. The full source code and specification are available on GitHub, and the author plans to increase difficulty if models improve.
Zscaler Report Details How Generative AI Accelerates Custom Malware Creation and Social Engineering
Zscaler published an analysis on October 16 examining the use of generative AI by attackers for social engineering, data discovery, and malware development. Principal security researcher Phil Hagen notes that the report focuses on broad methods rather than naming specific malware families or confirmed campaigns. The core observation centers on the dramatically increased speed at which malicious scripts can be written and iteratively revised. Traditional detection that relies on matching files or traffic against known patterns loses effectiveness when code is frequently altered while preserving the same functionality. The analysis clarifies that observed rapid creation and revision does not equate to fully autonomous self-mutating malware that changes on every execution. Another impact is the disappearance of classic indicators such as spelling and grammar errors in phishing messages, which previously helped identify malicious intent. Zscaler recommends focusing on behavioral analysis of what programs actually do rather than solely relying on matches to previously catalogued samples.