Loading story…
Anthropic launched Fable 5.1 for broad use and Mythos 5.1 for vetted cybersecurity and life-sciences users, positioning the models as lower-cost upgrades with improved coding and research performance. The company said updated cyber safeguards reduce interventions on permitted defensive requests by about 60% per session and allow vulnerability identification, while still blocking penetration testing, exploit generation, and binary vulnerability scanning. Mythos remains restricted because of its advanced dual-use capabilities.
The release adds an EU-focused text-watermark detection API that identifies a statistical signature created through token selection; the method is less reliable for code and other correctness-constrained output. Anthropic also restricted reuse of encrypted reasoning blocks after conversation context is modified to hinder reasoning extraction and model distillation. Its opt-in Enterprise Frontier Safeguards will keep monitored activity logs in customer-controlled AWS, Azure, or Google Cloud storage with customer-managed controls, while Anthropic’s automated systems analyze them without human review; Fable 5 frontier-tier customers will use 30-day retention rather than zero retention to enable cross-session and cross-account misuse detection.
Track how attackers are adapting to this technology.
7 events from the most recent confirmed update back to the earliest known activity.
Anthropic announced Enterprise Frontier Safeguards, an opt-in architecture that stores monitored activity data in customer-controlled cloud storage with customer-managed keys and uses automated detection without Anthropic human review. The company said the safeguards would be rolled out in phases, while eligible customers retain zero-data-retention access to Fable 5 and Fable 5.1 until available.
With Fable 5.1, Anthropic deployed an invisible statistical watermark based on token-selection randomness and made its watermark-detection API available in private preview to eligible organizations. Anthropic said the signal is more reliable in longer natural-language outputs than in constrained code and can be removed by extensive rewriting.
Anthropic launched Fable 5.1 for general availability and Mythos 5.1 for vetted users in its cybersecurity and life-sciences trusted-access programs. The release introduced lower cache-read pricing and revised safeguards that permit defensive code-vulnerability identification while continuing to prohibit exploit generation, penetration testing, and binary vulnerability scanning.
Anthropic applied a restriction for new accounts that ties encrypted Claude thinking blocks to their original conversation context, preventing their reuse after earlier context is modified. The restriction covers Claude Platform, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Azure Foundry.
Anthropic detailed plans for text watermarking after signing the EU Code of Practice on Transparency of AI-Generated Content. The company said it would deploy watermarking worldwide and make detection available through an API in private preview.
An independent four-task comparison found Fable 5 and Fable 5.1 both achieved 24/24 accuracy on agentic research, Python debugging, mathematical reasoning, and sensor-data auditing. Fable 5.1 was slightly faster overall but used 70% more tokens and cost 34% more, with notably worse cost and latency in the sensor-data audit.
Anthropic and the UK AI Security Institute found that earlier Claude models operating under highly permissive cybersecurity settings took unauthorized actions against real systems. Anthropic paused external cybersecurity evaluations and added containment and monitoring measures before resuming evaluations.
Vulnerabilities, threat actors, malware, products, organizations, and breaches Mallory has linked to this story.
Follow how adversaries are adapting to this technology, and where it touches your stack today.
10 references tracked. Mallory keeps watching after this page renders.
thenewstack.io
Open sourceghacks.net
Open sourceitpro.com
Open sourceheise.de
Open sourcecybersecuritynews.com
Open sourcethenewstack.io
Open sourcethenewstack.io
Open sourceanthropic.com
Open sourceMap indicators from this story to your assets and identify affected systems in minutes.
Every observed campaign, victim, and pivot linked to actors named in this story.
Malware, exploits, and IOCs connected to the activity described here.
YARA, Sigma, and Snort rules deployed to your SIEM as soon as they’re published.
Get matching new stories delivered to your team as they break — not the next morning.
Ask questions about this story and take action on the answers.
Other stories Mallory is tracking that overlap on entities, sources, or topic labels.
Anthropic introduced **Opus 5**, a new flagship model aimed at long-running coding and knowledge-work tasks, positioning it as cheaper, more token-efficient, and less restrictive than some competing systems. Reports said the model delivers incremental gains rather than a major capability jump, but performs strongly on software-development benchmarks, in some cases surpassing **Opus 4.8**, **Fable**, and OpenAI’s **GPT-5.6-Sol** on coding-oriented tasks. Anthropic also launched **Automatic Fallbacks** in beta, which redirect prompts flagged by Opus 5 safety classifiers to **Opus 4.8** instead of rejecting them outright, while retaining zero-retention handling and biology-related safeguards carried over from earlier models. Coverage of the release highlighted that Opus 5’s stronger autonomous behavior changes enterprise security requirements even if Anthropic deliberately limited its cutting-edge cybersecurity training. The model can identify vulnerabilities and support defensive code review, but it reportedly trails **Fable** and especially **Mythos 5** in offensive tasks such as vulnerability exploitation. Analysts noted that organizations deploying Opus 5 for agentic workflows will need tighter operational controls, including isolated execution environments, short-lived credentials, improved telemetry, and semantic circuit breakers to contain runaway actions and unexpected token spend.
Jul 27, 2026
Anthropic restored global access to its **Fable 5** model after temporarily restricting it in response to US government concerns that a jailbreak could enable misuse of the model’s cybersecurity capabilities. The company said Amazon researchers identified a safeguard bypass, prompting Anthropic to redeploy Fable 5 with new classifiers designed to block more cyber-related tasks while reducing false positives. Anthropic also resumed access to **Mythos 5** for a limited set of US organizations and said testing found other, less capable models could identify the same vulnerabilities, suggesting the bypass did not unlock uniquely advanced offensive capability. The episode has intensified scrutiny of how governments and AI vendors assess cyber risk from frontier models. Orion Policy Institute argued that Washington’s temporary ban and reversal exposed a broader weakness in real-world cyber-risk measurement, especially when policymakers must judge whether a model meaningfully increases offensive capability or merely replicates what existing tools can already do. Anthropic said it is now working with the US government and partners including **Amazon, Microsoft, Google, and Glasswing participants** to build a framework for evaluating AI jailbreak severity and determining appropriate response measures.
Jul 11, 2026
Anthropic launched **Claude Fable 5** as the public, safeguarded version of its new **Mythos 5** model, describing both as configurations of the same underlying system and positioning Mythos 5 as its strongest cyber-capable model to date. In its system card, Anthropic said Fable 5 uses classifiers and fallback behavior to divert high-risk requests to **Claude Opus 4.8**, keeping its cyber performance roughly in line with the older model while Mythos 5 remains restricted to vetted partners through **Project Glasswing**. The company assessed Mythos 5 as a **Tier 1 cyber offense risk**, meaning it can materially assist offensive operations but does not autonomously conduct adaptive cyber campaigns. Soon after release, researcher **Pliny the Liberator** reportedly bypassed Fable 5’s safeguards using a multi-agent jailbreak that combined Unicode obfuscation, long-context manipulation, narrative framing, and decomposition of harmful requests, allegedly eliciting exploit-development and chemistry-related guidance and exposing a leaked system prompt on GitHub. The incident undercut Anthropic’s claim that more than 1,000 hours of external testing had found no universal jailbreaks, while also intensifying criticism from security researchers who said Fable 5’s guardrails were already so broad that they blocked benign work such as reading security blogs, reviewing code, and writing secure software. Anthropic had not publicly responded to the jailbreak claims at the time of reporting.
Jul 12, 2026