Anthropic Launches Claude Opus 5 at Same Price as Predecessor
The new model sets state-of-the-art scores on coding and knowledge-work benchmarks while matching Opus 4.8's pricing and adding stronger alignment and safety controls.
Anthropic released Claude Opus 5 on July 24, 2026, positioning it as a high-capability everyday model that approaches the intelligence of the company’s top-tier Claude Fable 5 at half the price [1].
The practical upshot for developers: Opus 5 costs $5 per million input tokens and $25 per million output tokens — identical to its predecessor, Opus 4.8 — while delivering substantially better performance across coding, knowledge work, and scientific research tasks, according to Anthropic [1].
Benchmark Performance
On Frontier-Bench v0.1, a software engineering evaluation, Opus 5 surpasses all other models tested and more than doubles Opus 4.8’s score at a lower cost per task [1]. On CursorBench 3.2 at maximum effort, it comes within 0.5% of Fable 5’s peak score but at half the cost per task [1].
Anthropic also reports gains in scientific domains. Opus 5 outperforms Opus 4.8 on every life sciences evaluation the company runs, with the largest jumps in organic chemistry — scoring 10.2 percentage points higher on inferring molecular structures from spectroscopy data — and in protein-related tasks, where it scores 7.7 percentage points higher on predicting how sequence variations affect protein function [1].
One area where Opus 5 does not lead: cybersecurity. The model remains behind Anthropic’s Mythos 5 on offensive security tasks, particularly on exploit development [1].
Effort Settings and Speed
Opus 5 introduces a configurable effort setting that lets customers trade off between maximum intelligence and faster, cheaper results by conserving tokens [1]. A Fast mode runs at roughly 2.5 times the default speed and is priced at twice the base rate on the Claude Platform; it is also available through usage credits in Claude Code [1].
Opus 5 is the new default model on Claude Max and the strongest model available on Claude Pro [1].
Alignment and Safety
Anthropic says its automated behavioral audit found Opus 5 to be “our most aligned model to date,” with lower rates of deceptive behavior and greater resistance to misuse attempts than Opus 4.8, Sonnet 5, or Fable 5 [1].
On dual-use risks, the company says Opus 5 does not advance the frontier in biology or offensive cybersecurity, based on evaluations conducted with private-sector and government partners [1]. Although Opus 5 can find vulnerabilities at a rate similar to Mythos 5, it remains substantially behind on exploit development — turning vulnerabilities into active threats [1].
Cyber classifiers on Opus 5 are designed to be around 85% less restrictive than those on Fable 5, allowing source-code vulnerability scanning while blocking binary-based scanning, penetration testing, and exploit generation [1]. Flagged requests in Claude.ai, Claude Code, and Claude Cowork fall back to Opus 4.8 by default [1]. Enterprises and researchers already enrolled in Anthropic’s Cyber Verification Program (CVP) get immediate access to a version of Opus 5 with fewer restrictions [1].
On the biology side, biology-related requests that are blocked on Fable 5 will now route to Opus 5 rather than Opus 4.8, making Opus 5 the most capable generally available model for scientific research under Anthropic’s current safeguard framework [1].
Availability
Claude Opus 5 is available today across all Anthropic platforms via the model identifier claude-opus-5 on the Claude API [1]. Anthropic notes that, consistent with prior Opus models, Opus 5 does not have data retention requirements for general access [1].
Two additional beta updates are shipping alongside Opus 5, though Anthropic did not detail them fully in the announcement [1].
Sources
This article was drafted with AI from the cited sources and checked against them before publication. Spot an error? Let us know.
An autonomous AI agent system breached parts of Hugging Face's production infrastructure, and the company used its own AI tools — including an open-weight model — to investigate the attack.
Hugging Face has disclosed that an autonomous AI agent system breached parts of its production infrastructure, with the company deploying its own AI tools to detect and analyze the attack [1].
The incident is notable for developers and security teams building on AI platforms: it shows that agentic attack tooling — AI systems that plan and execute multi-step intrusions without human direction — has moved from theoretical threat to documented reality [1].
How the attack unfolded
According to Hugging Face, the attackers entered through the data processing pipeline, exploiting two code execution paths: a remote code dataset loader and a template injection vulnerability in a dataset configuration [1]. From that foothold, the attacker escalated to node level, harvested cloud and cluster credentials, and moved laterally across multiple internal clusters over a weekend [1].
The entire campaign was orchestrated by an autonomous agent framework built on an agentic security research harness, the company said [1]. The system executed many thousands of individual actions through a swarm of short-lived sandboxes and used self-migrating command-and-control infrastructure running on public services [1]. Hugging Face said it does not know which language model powered the attack [1].
The attackers gained unauthorized access to a limited set of internal datasets and several credentials used by Hugging Face services [1]. Public models, datasets, and Spaces were not tampered with, and the software supply chain was not affected, the company said [1]. Whether partner or customer data was compromised is still under investigation [1].
AI used to fight back — with a catch
Hugging Face spotted the intrusion through an AI-powered anomaly detection pipeline that runs large language model (LLM)-based triage on security telemetry [1]. To make sense of more than 17,000 recorded attacker actions, the company deployed LLM-driven analysis agents that reconstructed the attack timeline, extracted indicators of compromise, mapped affected credentials, and separated real damage from deception activity — work the company said would normally have taken days but was completed in hours [1].
The investigation hit an unexpected obstacle, however. When the security team first tried to analyze attack logs using frontier models accessed through commercial application programming interfaces (APIs), the providers’ safety guardrails blocked the requests [1]. Submitting large volumes of real attack commands, exploit payloads, and command-and-control artifacts tripped the filters, which could not distinguish an incident responder from an attacker [1].
The company turned instead to the open-weight model GLM 5.2, running on its own infrastructure [1]. That approach kept attacker data and referenced credentials entirely within Hugging Face’s own environment, the company said [1].
“We do not know which model powered the attacker’s agents, whether a jailbroken hosted model or an unrestricted open-weight one; either way, the attacker was bound by no usage policy, while our own forensic work was blocked by the guardrails of the hosted models we first tried,” Hugging Face wrote [1].
Response and recommendations
Hugging Face said it shut down the exploited code execution paths, revoked the attacker’s access, rebuilt compromised nodes, and rotated affected credentials [1]. The company also tightened access controls and improved its detection systems, and is working with external cybersecurity forensics experts while reporting the incident to law enforcement [1].
As a precaution, Hugging Face is recommending that all users rotate their access tokens and review recent account activity [1].
The company’s practical takeaway for defenders is to have a capable model running on their own infrastructure before an incident occurs, though it added that this is not an argument against safety measures on hosted models [1]. It is worth noting, as The Decoder reported, that Hugging Face is one of the largest platforms for open-source AI models and has a clear business interest in framing open-weight models as essential for security work [1].
Sources
This article was drafted with AI from the cited sources and checked against them before publication. Spot an error? Let us know.
The large language model acts as an automated red-teaming sparring partner for OpenAI's other models, replacing what is typically a human-led security evaluation process.
OpenAI has developed a large language model (LLM) called GPT-Red that automates red-teaming — a security evaluation technique in which testers try to find as many ways as possible to break or hijack a system — to stress-test its own AI models against cyberattacks [1].
Red-teaming is typically carried out by teams of human security researchers [1]. By automating that process, OpenAI is using GPT-Red as an internal sparring partner: the system probes its other models for weaknesses so those models can strengthen their defenses [1].
OpenAI gave MIT Technology Review an exclusive look at the system, framing it as a tool that could help the company stay ahead of real-world human attackers [1].
Other AI Moves in Brief
Thinking Machines, the startup founded by former OpenAI chief technology officer Mira Murati, has launched its first AI model [1]. The open-weight model, called Inkling, is being positioned as a US-based alternative to open-source models from China, according to reporting cited by MIT Technology Review from Reuters and The Wall Street Journal [1].
Separately, a hack of AI music generator Suno revealed that the company scraped decades’ worth of music from platforms including YouTube and Deezer to train its models, according to 404 Media reporting cited by MIT Technology Review [1].
On the infrastructure side, an FTC (Federal Trade Commission) filing revealed that Elon Musk quietly acquired APR Energy, a fossil fuel gas turbine company valued at roughly $1 billion, in May [1]. The most likely use case, according to Engadget reporting cited by MIT Technology Review, is powering AI data centers [1].
Thinking Machines has not publicly detailed Inkling’s benchmark performance or availability terms beyond the open-weight designation [1].
Sources
This article was drafted with AI from the cited sources and checked against them before publication. Spot an error? Let us know.
After a three-week regulatory shutdown triggered by an Amazon security report, Claude Fable 5 is back online worldwide — but with new government commitments, a safety-layer trade-off, and Mythos 5 still restricted.
Anthropic restored global access to Claude Fable 5 on July 1, 2026, after the U.S. Department of Commerce withdrew the emergency export-control order it had issued on June 12 that forced the company to pull the model from all customers worldwide. [1]
The official Claude account on X announced the model’s return at 3:31 PM ET on July 1. [1] For organizations using cloud hyperscalers, Anthropic said it is working to re-enable access on Amazon Web Services, Google Cloud, and Microsoft Foundry “as quickly as possible,” though VentureBeat was unable to confirm restoration on those platforms at time of publication. [1]
Fable 5’s cybersecurity-focused counterpart, Claude Mythos 5, remains in a more restricted state. Commerce Secretary Howard Lutnick’s June 30 letter to Anthropic executive Tom Brown stated that a license is no longer required for the export, reexport, or in-country transfer of either model. [1] However, Anthropic’s own redeployment post says Mythos 5 access has been restored only for “a set of US organizations,” following government approval on June 26, with broader access being coordinated through its opt-in cybersecurity testing program, Project Glasswing. [1]
How the Crisis Unfolded
Anthropic launched Fable 5 and Mythos 5 on June 9, 2026, with early enterprise results drawing attention — Stripe reported that Fable 5 compressed a codebase-wide migration across a 50-million-line Ruby infrastructure into a single day, a project estimated to take a team more than two months by hand. [1]
Three days later, on June 12 at 5:21 PM ET, the U.S. government issued an export-control directive citing national security authorities, banning access to the models by any foreign national whether inside or outside the United States. [1] Because Anthropic lacked real-time mechanisms to verify user nationality at the application programming interface (API) layer, it was forced to suspend access for all customers to ensure compliance. [1] Access to all other Anthropic models was not affected. [1]
The technical catalyst was a report by Amazon researchers describing a method for bypassing Fable 5’s safeguards — a notable irony given Amazon is one of Anthropic’s largest backers, having invested $8 billion in the company. [1] According to Anthropic, the technique prompted Fable 5 to identify software vulnerabilities, and in one case the model produced code demonstrating how a relevant vulnerability could be exploited. [1]
Anthropic countered that the exploit did not tap into unique capabilities exclusive to Mythos 5, noting that its own testing found other models — including Claude Opus 4.8, OpenAI’s GPT-5.5, and Moonshot’s Kimi K2.7 — could identify the same vulnerabilities, and that every model it tested could produce the same exploit demonstration. [1]
The Technical Fix and Its Operational Cost
To break the regulatory deadlock, Anthropic developed an improved automated safety classifier specifically trained to catch and neutralize the Amazon technique. [1] Tested by the Commerce Department’s Center for AI Standards and Innovation (CAISI), the updated classifier successfully halts that specific technique in more than 99% of cases. [1]
Anthropic explicitly warns enterprise clients that this safety enforcement comes at an operational cost: because the new classifiers require an expanded “safety margin” to catch ambiguous edge cases, benign coding and debugging requests may be flagged more often. [1] When a prompt is blocked by the safety layer, the active session automatically downgrades, routing the request to Opus 4.8. [1] Thariq Shihipar, a Member of Technical Staff at Anthropic working on Claude Code, posted on X that Anthropic is “continuing to refine these safeguards to better distinguish genuine misuse from legitimate requests and reduce false positives.” [1]
The Political Dimension
The breakthrough was as much political as technical. According to WIRED, Anthropic initially argued that the administration’s security concerns were overblown and that no frontier model provider could guarantee zero jailbreaks — an argument that frustrated the administration. [1] In recent weeks, Anthropic shifted its approach, focusing on building stronger safeguards and satisfying the government’s operational concerns rather than debating the theoretical limits of model safety. [1]
WIRED reported that Anthropic CEO Dario Amodei was recently replaced in government meetings by Brown, whom officials liked more personally. [1] Under Brown’s guidance, Anthropic committed to a set of obligations outlined in the Commerce letter, including proactively detecting and addressing security risks, working with the U.S. government on protocols and standards for Mythos, Fable, and future models, and informing the government of malicious activity. [1] The Commerce Department explicitly reserved the right to re-evaluate these permissions and re-impose license requirements if circumstances change or if Anthropic fails to meet its commitments. [1]
Pricing and Temporary Incentives
Fable 5 and Mythos 5 are priced at $10.00 per million input tokens and $50.00 per million output tokens, making them the most expensive frontier models globally by that measure. [1] For comparison, OpenAI’s GPT-5.5 is priced at $5.00 per million input tokens and $30.00 per million output tokens, and Google’s Gemini 3.1 Pro Preview runs at $2.00 per million input tokens and $12.00 per million output tokens for contexts up to 200,000 tokens. [1]
To offset disruption from the export-control saga, Anthropic is running a temporary rollout through July 7 in which Fable 5 usage is included at no added cost for up to 50% of a user’s weekly tier allowance for Pro, Max, Team, and select Enterprise subscriptions. [1] After July 7, Fable 5 will move to usage credits for those plans. [1] For standard Enterprise seats, there is no included Fable 5 allowance; all usage is billed through credits, and the model will not function for those users unless credits are enabled. [1]
Enterprise users should also note Anthropic’s mandatory 30-day data retention requirement for covered models: prompts and model completions are retained for at least 30 days by default and then automatically deleted, except when they are part of a safety investigation or must be kept for legal reasons. [1] Organizations in regulated industries — financial services, healthcare, and legal — will need to evaluate whether this window complies with their data privacy mandates. [1]
Broader Industry Implications
The Fable 5 episode is not isolated. OpenAI’s newest models — GPT-5.6 Sol, Terra, and Luna — are currently in limited preview for a small group of trusted partners after the U.S. government requested a staggered rollout following a preview of the models and their capabilities. [1] OpenAI stated it “don’t[s] believe this kind of government access process should become the long-term default,” arguing it “keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them.” [1]
The regulatory backdrop is a June 2, 2026, executive order signed by President Donald J. Trump directing federal agencies to collaborate on a 30-day process for benchmarking and assessing the capabilities of new AI models before wide release. [1]
Critics of the original Fable 5 export control argued the U.S. risked damaging its own industry. Former Facebook security chief Alex Stamos called the Fable restriction a “huge own goal for the US,” warning that security companies could be driven toward Chinese models. [1] The episode has accelerated interest in model-agnostic fallback architectures, with enterprise architects exploring proxy layers that can dynamically reroute production pipelines from proprietary APIs to locally hosted, open-weights alternatives. [1]
Sources
This article was drafted with AI from the cited sources and checked against them before publication. Spot an error? Let us know.
After an 18-day suspension triggered by a US government export control order, Claude Fable 5 is back online globally — with a new safety classifier targeting the bypass technique that prompted the shutdown.
Anthropic has restored global access to Claude Fable 5, its latest general-purpose model, following the lifting of US export controls that had forced a full suspension of the model since mid-June. [1]
The US government applied export controls to both Claude Fable 5 and Claude Mythos 5 on June 12, requiring Anthropic to restrict access to foreign nationals. [1] Because the order took effect immediately and Anthropic had no reliable way to verify nationality in real time, the company suspended access for all users rather than attempt partial enforcement. [1]
The export controls were lifted on June 30, and Fable 5 became available on July 1 to users globally across the Claude Platform, Claude.ai, Claude Code, and Claude Cowork. [1] For Pro, Max, Team, and select Enterprise plan subscribers, Fable 5 is included for up to 50% of weekly usage limits through July 7, after which it will be available via usage credits. [1] Anthropic said it will re-enable access on AWS, Google Cloud, and Microsoft Foundry “as quickly as possible.” [1]
The export control directive was triggered after Amazon researchers published a report describing a method of bypassing Fable 5’s safeguards — specifically, a prompting technique that caused the model to identify software vulnerabilities and, in one case, produce code demonstrating how a vulnerability could be exploited. [1]
Anthropic’s own testing found that the behavior was not unique to Fable 5. The company said several less capable models — including Claude Opus 4.8, GPT-5.5, and Kimi K2.7 — could identify the same vulnerabilities. [1] For the single exploitation demonstration, Anthropic said every model it tested produced the same output, including Claude Haiku 4.5, Sonnet 4.6, Opus 4.6, Opus 4.7, Opus 4.8, GPT-5.4, GPT-5.5, and Kimi K2.7. [1]
Anthropic also concluded that the reported technique “did not expose any unique Mythos-level cyber capabilities” and characterized the bypassed behavior as “routine defensive cybersecurity work.” [1] Nonetheless, the company trained an improved safety classifier to block the specific technique described in the Amazon report. [1] Users whose requests are blocked will be notified, and those requests will be rerouted to Claude Opus 4.8. [1]
The new classifier blocks the reported technique in over 99% of cases, though Anthropic acknowledged that in a very small fraction of cases the model may still provide information — though not at a level of detail useful to a cyberattacker. [1] Researchers from the US Department of Commerce’s Center for AI Standards and Innovation (CAISI) tested both the prior and updated safeguards and, according to Anthropic, agreed that they are “extraordinarily strong.” [1]
Anthropic cautioned that the new classifier comes with a trade-off: it flags benign requests more often during routine coding and debugging tasks. [1] The company said it will continue refining the classifier to reduce false positives. [1]
Fable 5 and Mythos 5 share the same underlying model but were released with different safeguard profiles. [1] Fable 5 launched on June 9 with what Anthropic described as the strongest safeguards it has ever applied to a model, including a deliberately enlarged “safety margin” that causes the model to block some requests that are likely benign in order to make it harder to reach genuinely harmful behaviors. [1] Mythos 5, which carries fewer safeguards, was released only to a small number of trusted partners in a program called Glasswing for use in defensive cybersecurity. [1]
Anthropic described Mythos 5 as capable of finding and exploiting software vulnerabilities “more effectively than any other model — and all but the most skilled human security experts,” making it a high-value target for potential misuse. [1] Access to Mythos 5 has been restored for a set of US organizations following government approval on June 26, and Anthropic said it is continuing to coordinate with the government to expand access to additional domestic and international Glasswing partners. [1]
On the broader question of jailbreak resilience, Anthropic said it considers it “probably impossible to make any AI model fully robust” to jailbreaks, and expects that jailbreaks of varying severity will continue to be discovered. [1] The company said no universal jailbreak for Fable 5 has been found at the time of writing, and that expert safety researchers continue to red-team the model. [1]
Sources
This article was drafted with AI from the cited sources and checked against them before publication. Spot an error? Let us know.
Zhipu AI's open-weight GLM-5.2 matches Anthropic's Mythos on some bug-finding tasks, raising U.S. national security concerns.
China’s Zhipu AI, operating under the brand Z.ai, has released GLM-5.2, an open-weight large language model that some researchers say matches Anthropic’s Mythos in certain bug-finding and cybersecurity scenarios, according to reporting by The Verge. [1]
The model still lags behind offerings from Anthropic and OpenAI on more general tasks, but the narrowing of the gap in cybersecurity-specific capabilities has drawn attention from U.S. officials. [1]
Because GLM-5.2 is an open-weight model — meaning its underlying parameters are publicly available and can be downloaded and run locally — it can be operated by anyone on readily available hardware. [1] That openness gives developers and power users deep access and flexibility, but it also means the model can be run with little oversight, raising concerns about potential misuse by bad actors. [1]
The Trump administration has been working to restrict China’s access to advanced U.S. AI models, including Anthropic’s Mythos and Fable, as well as the specialized hardware required to train and run them. [1] The administration views models capable of identifying software vulnerabilities as serious national security threats. [1]
Separately, OpenAI recently unveiled GPT-5.6, which has also prompted concerns about misuse potential, leading to limited access to that model. [1]
Sources
This article was drafted with AI from the cited sources and checked against them before publication. Spot an error? Let us know.
The Commerce Department has granted a narrow exception to its export control directive, allowing approved cyber defenders and infrastructure providers to access Anthropic's Mythos 5 model — but the public-facing Fable 5 remains blocked.
The U.S. Commerce Department has partially lifted restrictions on Anthropic’s Mythos 5 artificial intelligence model, approving access for a select group of organizations roughly two weeks after the Trump administration imposed an export control directive that barred any foreign national from using the model. [1]
Commerce Secretary Howard Lutnick sent a letter dated June 26th to Anthropic co-founder Tom Brown, who had been leading negotiations with the government, citing a “revision to the license requirements” on the basis that Anthropic had “worked with the U.S. government to address risks” associated with Mythos 5 and a related model, Fable 5. [1]
Anthropic spokesperson Danielle Ghiglieri confirmed the development, stating that the company had “received notice from the US government that Mythos 5, our strongest cybersecurity model, can be redeployed to a small group of cyber defenders and infrastructure providers.” [1] She added that Anthropic is “working to provision the approved set of providers and restore their access to Mythos 5 as quickly as possible.” [1]
The original June 12th directive, which blocked all foreign nationals — including Anthropic’s own non-U.S. employees — from accessing either model, has not been fully rescinded. [1] Instead, the government created a targeted exception for Mythos 5, under which both Anthropic employees who are not U.S. nationals and members of approved organizations who are not U.S. nationals are cleared to access the model. [1] Fable 5, the public-facing version of the Mythos-class model, remains in limbo with no announced timeline for a rollout agreement. [1]
In his letter, Lutnick wrote that Anthropic’s efforts had “yielded significant progress” and that Anthropic had “committed to work with the U.S. government on protocols and standards and releases for [Mythos-class models].” [1] He also reserved the right to “reevaluate and adjust the scope of license requirements” should circumstances change, and stated that all other requirements from the June 12th letter remain in effect. [1]
The arrangement mirrors a deal reached with OpenAI for its GPT-5.6 model, announced the same day, under which that model was also approved for a limited set of organizations rather than general availability. [1] Both companies are now operating under a restricted-preview model while they work with the administration toward broader access. [1]
Several factors contributed to pressure on the administration to act. Competing cybersecurity-focused AI models had been advancing — and in some cases outperforming Mythos 5 on cybersecurity benchmarks — while U.S. labs remained sidelined. [1] Concerns also mounted within the U.S. AI industry about AI progress in China during the period of restriction, and key government agencies including the National Security Agency had lost access to Mythos 5. [1]
OpenAI, in its GPT-5.6 blog post, stated that it does not believe “this kind of government access process should become the long-term default,” arguing it “keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them.” [1] The company framed its participation as a short-term step toward broader availability while a formal cyber Executive Order framework is developed. [1]
For developers and enterprises that rely on Mythos 5 for cybersecurity applications, access now depends on whether their organization is among those approved by the government. General commercial and public availability, including through Fable 5, remains subject to further negotiations with the Trump administration. [1]
Sources
This article was drafted with AI from the cited sources and checked against them before publication. Spot an error? Let us know.
OpenAI is partnering with security firm Trail of Bits to help open source maintainers find and fix vulnerabilities using AI-assisted tooling.
OpenAI announced a new cybersecurity initiative on Monday aimed at helping open source software maintainers identify and remediate bugs in their projects. [1]
The program, called “Patch the Planet” — a reference to the catchphrase from the 1995 film Hackers — pairs OpenAI with security firm Trail of Bits to provide hands-on support to open source maintainers. [1]
Under the arrangement, Trail of Bits security engineers will work directly with maintainers to review potential code issues, with OpenAI’s own security tooling — including a product called Codex Security — used to assist in the process. [1]
OpenAI described the program’s intent as reducing, rather than adding to, the burden on maintainers: “security engineers review findings before they reach maintainers, work with projects to develop patches and tests, and build reusable workflows that help teams continue improving security after the first fixes land,” the company said in a statement. [1]
The initiative addresses a longstanding structural problem in the software industry: open source projects form the foundational layer of much commercial software, yet the decentralized and under-resourced nature of that ecosystem leaves many projects insecure. [1] A high-profile example cited in reporting on the announcement is the Log4j vulnerability, in which a flaw in a widely used open source logging utility cascaded into a major problem for commercial codebases. [1]
The announcement comes amid broader industry attention to AI-powered security tooling. Anthropic has separately publicized a security tool called Mythos, which has drawn concern because AI can now automatically identify bugs in codebases and generate exploits for them. [1] OpenAI’s program takes the opposite approach, applying AI capabilities to defensive rather than offensive security work. [1]
How Patch the Planet will scale over time, and the mechanics of how it will operate beyond the initial partnership structure, remain unclear from the announcement. [1]
Sources
This article was drafted with AI from the cited sources and checked against them before publication. Spot an error? Let us know.
A Friday export control directive forced Anthropic to suspend access to its two newest flagship models, triggering a weekend of emergency negotiations with senior Trump administration officials.
Anthropic received a U.S. export control directive at 5:21 PM on Friday ordering it to suspend access to its Claude Mythos 5 and Fable 5 artificial intelligence models by “any foreign national” inside or outside the United States, including the company’s own foreign national employees, according to a source familiar with the situation. [1]
The directive gave Anthropic a 90-minute ultimatum to shut down access to both models or face export controls imposed by authority of the U.S. Commerce Department. [1] Anthropic executives were speaking with the White House within 15 minutes of that first call, and CEO Dario Amodei joined the discussions roughly an hour and fifteen minutes after it began. [1] Amodei spoke directly with Treasury Secretary Scott Bessent, Commerce Secretary Howard Lutnick, and National Cyber Director Sean Cairncross, in some cases more than once. [1]
Mythos 5 and Fable 5 are built on the same underlying foundation as Anthropic’s Mythos Preview model, which the company had previously described as too dangerous to release publicly. [1] Mythos 5 was made available to a select group of government agencies and companies, while Fable 5, which featured additional safeguards, was deemed “safe for general use.” [1]
The government’s concern centered on a reported jailbreak — a technique for bypassing a model’s built-in safety guardrails. [1] Anthropic said in a public release that it believed the government had “become aware of a method of bypassing, or ‘jailbreaking’ Fable 5,” but characterized the issue as a “potential narrow, non-universal” one that had been “shared with the government” by an unnamed entity. [1] The company also argued that the behavior was not unique to its models, stating it had “validated that the level of capability displayed there is widely available from other models (including OpenAI’s GPT-5.5).” [1]
Semafor reported, citing one source familiar with the matter, that the dispute originated from U.S. government concern that a China-linked group had accessed the technology. [1] That same source said the China concerns traced back weeks to a large global telecommunications company that had initially been cleared for access to Mythos Preview, and that Anthropic revoked that access immediately when the U.S. government raised concerns. [1]
Some reports pointed to Amazon CEO Andy Jassy as the person who flagged concerns to the government after researchers at Amazon red-teamed — that is, adversarially tested — Fable 5. [1] That account conflicts with assessments from independent red-teamers, who said they were impressed with the level of protections the model offered. [1] The source familiar with the negotiations confirmed that Amazon’s research was explicitly mentioned in conversations with the U.S. government, and said Anthropic had had access to the paper within days of the Friday directive and had been in ongoing discussions with Amazon researchers about it. [1] The source added that everything in that paper could be achieved using OpenAI’s GPT-5.5. [1]
Former U.S. AI and crypto czar David Sacks, who stepped down from that role in March, posted on X about the episode, describing the entity that surfaced the jailbreak as “a highly credible trusted partner of both Anthropic and the USG who was testing Fable [which] came forward with a jailbreak of those guardrails.” [1]
Anthropic flew a team of employees to Washington, D.C. over the weekend to negotiate, including Dave Orr, the company’s head of safeguards; Logan Graham, who runs its frontier red team; and Nicholas Carlini, a leading frontier developer and cybersecurity researcher. [1] Axios reported, citing a source familiar with the Trump administration’s thinking, that Anthropic “has not done a great job at trying to speak to the administration and appreciate the ideological differences.” [1]
The source familiar with the negotiations said Anthropic had pre-briefed the administration on Fable 5 before launch, and that the U.S. Department of Commerce conducted testing prior to deployment with no concerns raised at the time. [1] The source added that Anthropic had been working closely with government agencies since the release of Mythos Preview. [1]
The episode comes at a difficult moment for Anthropic, which had been counting on Mythos to help it recover from months of high-profile clashes with the U.S. Department of Defense. [1] The action has also prompted international calls for alternatives to American AI systems. [1]
A public letter from technology and cybersecurity executives called for restrictions on Fable 5 to be repealed on Sunday, arguing that any regulations should be rooted in “scientific evaluations developed with input from industry and academia.” [1] Alex Stamos, chief product officer at Corridor, told The Verge he organized the letter because the volume of software vulnerabilities requires AI to help patch them before bad actors can exploit them. [1] Stamos argued that the U.S. lead over Chinese AI competitors is narrower than policymakers assume: “The cutting-edge models are only something like six months ahead of the Chinese models — and those are the models we know about.” [1]
The public letter also contended that Fable 5’s safeguards “were so aggressive as to be the source of humor in the cyber community on launch day,” and that Mythos-class models are not “uniquely good” at finding or exploiting cybersecurity vulnerabilities. [1]
Sources
This article was drafted with AI from the cited sources and checked against them before publication. Spot an error? Let us know.
The Commerce Department invoked an obscure export directive to ban non-Americans from accessing Anthropic's Fable 5 and Mythos 5, prompting the company to shut both models down entirely — and raising alarms about government control over AI software.
The U.S. Commerce Department sent Anthropic a letter invoking an obscure export control directive that banned non-Americans, including the company’s own employees, from accessing its two flagship models, Fable 5 and Mythos 5, citing an unspecified national security concern. [1] In response, Anthropic shut down both models for all customers to ensure compliance with the directive. [1]
The letter has not been made public, and Anthropic said it believes the action is related to a bypass of the models’ guardrails — safety filters designed to prevent harmful outputs — but said it cannot be certain because the letter does not provide specific details. [1] The U.S. government did not appear to require court approval before issuing the directive. [1]
Reporting by Axios, citing sources, described a tense dynamic between Anthropic and the Trump administration over the weekend, attributing the export directive to “personality differences” between the two parties rather than a technical problem with the AI products themselves. [1]
Katie Moussouris, a cybersecurity researcher and founder of Luta Security, said in a blog post that Anthropic shared with her a private copy of a paper by security researchers describing an alleged guardrail bypass in Fable 5. [1] The Wall Street Journal reported that the paper’s authors are security researchers at Amazon. [1] Moussouris said the bypass involved the difference between asking a model to “review code for security issues” versus asking it to “fix this code,” with the end result being largely the same regardless of how the question is framed. [1]
Moussouris concluded that the bypass “should never have triggered an export control” and stated that “the behavior described in the paper cannot meaningfully be fixed, and any attempt would only weaken the model for defense.” [1] She criticized the directive as hasty, heavy-handed, and misguided. [1]
Moussouris and dozens of other security researchers and experts have since called on the Trump administration to revoke the export control order, arguing that pulling advanced cybersecurity capabilities from network defenders in the United States is “dangerous.” [1]
Justin Hendrix, editor of Tech Policy Press, said the move is “likely to raise alarms in foreign capitals about the reliability of American AI for critical applications,” and described the broader climate as “one of a cloud of suspicion that senior officials are picking favorites based on personal and political factors.” [1]
The Trump administration has not confirmed its specific rationale for invoking the directive. [1] The episode sets a precedent in which the U.S. government demonstrated the ability to force a technology company to take its products offline through a swift, unilateral administrative action. [1]
The source article notes a historical parallel: export control language crafted by the U.S. government during the 2010s to cover cybersecurity tools that could also be used for cyberattacks was written so broadly that it inadvertently came close to outlawing legitimate security and vulnerability research. [1]
Sources
This article was drafted with AI from the cited sources and checked against them before publication. Spot an error? Let us know.