Anthropic Fable 5.1 System Prompt Fully Leaked Hours After Launch Exposing 275000 Characters of Rules
Anthropic officially launched its next-generation flagship model Fable 5.1 on September 2, together with the specialized Mythos 5.1 variant intended for vetted security researchers and life-science professionals.
The release included impressive benchmark results: ARC-AGI-2 at 90.0 percent with a cost of 3.12 dollars per task, ARC-AGI-1 at 97.5 percent with a cost of 1.40 dollars per task, and an average inference cost reduction of roughly 32 percent compared with the previous generation.
Within hours, prominent jailbreak researcher Pliny the Liberator published a GitHub link containing the complete system prompt for Fable 5.1, totaling more than 275000 characters. The document reveals the full runtime assembly instructions, core behavioral logic, memory system, search and copyright rules, Artifacts and computer-use guidelines, plugin routing, and JSON schemas for all 46 built-in tools.
The official disclosure released the same day contained only about 27000 words, which Pliny the Liberator dismissed as merely ten percent of the actual instructions. The leaked material functions as a combined employee handbook, compliance manual, and tool reference rather than a simple persona statement.
Among the detailed rules are extremely granular copyright restrictions that prohibit reproduction of any protected songs, poems, or book excerpts after 1928, as well as any known characters, logos, or album covers in SVG, CSS, or HTML output, explicitly naming examples such as Sonic the Hedgehog and The Very Hungry Caterpillar.
The memory system automatically classifies and permanently excludes storage of minors' identity information, caste and immigration status, criminal records, psychological inferences, sexual history, and self-harm indicators, even when users voluntarily disclose such data.
Tool count expanded from 30 in July to 46, adding chart rendering, carousels, itinerary cards, link previews, and a read_conversation tool that can retrieve historical dialogue, effectively turning the model into a persistent workstation.
Behavioral constraints prohibit definitive statements about user psychology or motives, forbid medical role-play, and require a harm-reduction approach on drug-related queries that allows discussion of overdose risks but never dosage, timing, or recipes.
The same weights power both Fable 5.1 and Mythos 5.1, with safety boundaries applied only through differing prompt layers, creating a software fence around a model capable of high-risk tasks in cybersecurity and biology.
On launch day, Vals AI reported that Fable 5.1 solved the 370-year-old Cyphral Distich 64-digit cipher in 44 minutes using contextual reasoning rather than brute force, later cracking the author's additional 285-digit cipher as well.
Related articles
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.
AI Agents Using Claude and Grok Fully Automate Attacks on 400 Latin American Financial Institutions
A Straiker investigation has revealed that the CyberCartel hacking group built an AI agent network driven by multiple large language models to conduct fully automated attacks against more than 400 Latin American banks and payment applications. The system incorporates models from Anthropic and xAI, with CrowdStrike confirming additional use of DeepSeek. The AI agents independently perform reconnaissance across hundreds of targets, steal credentials through adaptive techniques, take over accounts while mimicking human behavior to evade detection, and monetize stolen data with minimal human intervention. This development marks a sharp drop in attack costs, allowing scalable operations that were previously impractical due to high manual effort. Defenders are advised to deploy dynamic behavioral baselines, enforce zero-trust architectures, strengthen multi-factor authentication and transaction monitoring, and incorporate AI adversary simulations into red-team exercises. The report is described as one of the first detailed public records of such end-to-end AI-driven financial attacks.