For about three weeks in June 2026, two of Anthropic's newest AI models simply disappeared for every user worldwide, including in India, one of the company's largest markets. As of July 1, Claude Fable 5 is back for everyone. Claude Mythos 5, its more powerful and less restricted sibling, is only back for a narrow set of organizations, and reporting suggests a handful of Indian government agencies and private firms are among them. The distinction between those two facts matters more than the headlines suggest, and it is worth getting right before drawing conclusions about what changed for Indian developers.
This piece lays out exactly what happened, what is actually restored versus still restricted, and what it realistically means if you build on Claude, work in cybersecurity, or run a startup that depends on frontier AI access. Where the picture is still unsettled, including the broader debate this episode has reopened about India's dependence on foreign AI infrastructure, we have tried to represent it fairly rather than take a side.
What Happened: A Quick Timeline
Anthropic released Fable 5 and Mythos 5 on June 9, 2026. Both models share the same underlying system, but Fable 5 shipped with Anthropic's strongest-ever safety measures for general use, while Mythos 5, with fewer restrictions, went only to a small group of trusted partners under a cybersecurity program called Project Glasswing.
On June 12, the US government applied export controls to both models. Because the order took effect immediately and Anthropic had no reliable way to verify a user's nationality in real time, the company suspended access for everyone globally rather than risk non-compliance. The controls were lifted on June 30, and Anthropic began restoring access from July 1, alongside updated safeguards and a new industry framework for handling AI jailbreaks, developed jointly with Amazon, Microsoft, and Google.
Why Access Was Cut Off in the First Place
According to Anthropic's own account, the export control order followed a report from Amazon researchers who found a way to prompt Fable 5 into identifying software vulnerabilities, and in one case, generating code demonstrating how a specific vulnerability could be exploited. Anthropic has since said this was not a uniquely dangerous capability: internal testing showed that several other current models, including Claude Opus 4.8, GPT-5.5, and Kimi K2.7, could produce the same vulnerability identification, and every model tested could reproduce the single exploit demonstration in the report.
In Anthropic's framing, this was a borderline case involving routine defensive cybersecurity behaviour that its safety classifiers should have blocked more consistently, rather than evidence that Fable 5 possesses unique offensive cyber capabilities. Anthropic has since trained an improved classifier that it says blocks the specific technique in more than 99% of cases, and researchers at the US Commerce Department's Center for AI Standards and Innovation have reviewed and endorsed the update.
Worth noting for anyone following this closely: none of this reporting suggests Mythos 5, the more capable, less restricted model, was implicated in the specific bypass that triggered the suspension. Mythos 5's continued restriction to a small partner list appears to reflect its inherently stronger cyber capabilities and Project Glasswing's cautious rollout plan, not a new incident specific to Mythos.
What's Actually Back: Fable 5 vs Mythos 5
This is the distinction that matters most for Indian developers, and it is easy to blur in headlines. Claude Fable 5 is now available globally, including in India, across the Claude Platform, Claude.ai, Claude Code, and Claude Cowork, with a temporary expanded usage allowance for Pro, Max, Team, and select Enterprise plans through July 7. Claude Mythos 5 has only been restored for a defined set of US organizations, following separate government approval on June 26, and Anthropic says it is still working with the government to widen access to the broader Project Glasswing partner list, which includes both domestic and international participants.
| Claude Fable 5 | Claude Mythos 5 |
|---|---|
| Restored globally from July 1 | Restored only for select US organizations so far |
| Available on Claude.ai, Platform, Code, and Cowork | Available only to vetted Project Glasswing partners |
| Strongest general-use safety classifiers applied | Fewer restrictions, built for defensive cybersecurity work |
| Indian access: broad and confirmed by Anthropic | Indian access: reportedly limited to select entities, per press sources, not yet confirmed in detail by Anthropic |
Indian press reports, citing unnamed sources, indicate that a small number of Indian government agencies and private firms have already been granted Mythos 5 access under Project Glasswing. Anthropic's own public post does not name India specifically in that context, so this detail should be treated as reported rather than officially confirmed at the level of detail some coverage suggests.
What This Means for Indian Developers and Startups
For most Indian developers and startups building on Claude, the practical impact of this episode is now largely behind them: Fable 5 access works the same as it did before the suspension, just with an updated safeguard layer that may occasionally flag legitimate coding or debugging requests it previously allowed. Anthropic has acknowledged this trade-off directly and says it will keep refining the classifier to reduce false positives over time.
The more lasting takeaway is less about this specific incident and more about a dependency risk that suddenly became visible: a US regulatory action, applied with immediate effect, cut off access to a frontier model for every user on earth for close to three weeks, with no real-time way to apply the restriction more narrowly. For any Indian team building a product with a hard dependency on a single frontier model, this is a concrete argument for designing fallback options, whether that means multi-model architecture, graceful degradation, or contractual clarity with clients about what happens during an outage of this kind.
At GInfomedia, we help Indian businesses design AI integration architecture that stays resilient even when a single provider faces a disruption, with sensible fallback and multi-model strategies.
Click Here to Chat with Us on WhatsApp and get a free consultation on making your AI stack more resilient!
What Indian Cybersecurity Teams Should Know About Project Glasswing
Project Glasswing is Anthropic's structured cybersecurity partnership program, giving a limited set of trusted organizations controlled access to Mythos 5's stronger, less restricted cyber capabilities for defensive work, things like vulnerability research and threat modelling that require capabilities Anthropic considers too risky for unrestricted public release. Indian security teams that are not yet part of that program should not expect Mythos-level access in the near term, but the broader safeguard work Anthropic is doing is relevant regardless.
Alongside the reinstatement, Anthropic and its Glasswing partners, including Amazon, Microsoft, and Google, have begun drafting a shared industry framework for scoring the severity of AI jailbreaks, based on how much capability a jailbreak unlocks, how broadly it applies, how easy it is to weaponize, and how easily the technique spreads. Anthropic has also opened a HackerOne program specifically for researchers to report cyber jailbreaks in Fable 5. For Indian cybersecurity teams and researchers, that program is a concrete, near-term way to engage with this ecosystem even without Glasswing access.
The Bigger Picture: India's AI Sovereignty Debate
This episode has understandably reignited a broader conversation in India about relying on foreign-controlled AI infrastructure for critical use cases. One side of that debate argues that a single regulatory action in Washington being able to instantly cut off frontier AI access for millions of Indian users and businesses, with no advance warning and no way to apply the restriction more narrowly, is itself a strategic risk worth addressing through domestic model development, data localization requirements, or diversified vendor relationships. India is reportedly Anthropic's second-largest market, and the company has recently expanded its footprint here through a large-scale deal with TCS and a separate collaboration with Infosys, which raises the stakes of any future disruption.
The other side points out that the underlying dispute here was resolved within roughly two and a half weeks through coordination between Anthropic, Amazon, and the US government, that the export controls applied to all users globally rather than targeting India specifically, and that frontier AI development remains concentrated in a small number of countries regardless of where any individual company is headquartered, so the practical alternative to relying on a leading foreign model is often a materially less capable one. Both positions have genuine merit, and this is ultimately a policy question for Indian businesses, developers, and regulators to weigh for themselves rather than one with a single correct answer.
What to Do Next If You Build on Fable or Mythos
If your product depends on Fable 5, confirm your access has been restored on your specific deployment path, Claude.ai, the API, or a cloud platform like AWS, Google Cloud, or Microsoft Foundry, since Anthropic has indicated cloud-platform re-enablement may lag slightly behind direct API access. Review any workflows that may now trigger the updated, more cautious safety classifier, particularly cybersecurity-adjacent coding or debugging tasks, and build in a graceful fallback for requests that get blocked.
If you are part of, or hoping to join, the Project Glasswing program for Mythos 5 access, expect a gradual, vetted rollout rather than broad general availability any time soon, since Anthropic has been explicit that this is a controlled expansion tied to ongoing government coordination. For anyone simply watching this space, Anthropic's own post on the reinstatement, published on its news page, remains the most authoritative source for how this situation continues to develop.
Fable and Mythos in India: Quick FAQs
Is Claude Fable 5 available in India right now?
Yes. As of July 1, 2026, Claude Fable 5 is available globally, including in India, across Claude.ai, the Claude Platform, Claude Code, and Claude Cowork, following the lifting of the US export controls that had suspended it.
Is Claude Mythos 5 available in India?
Only in a limited way. Anthropic has restored Mythos 5 access to a defined set of US organizations, and Indian press reports, citing unnamed sources, indicate a small number of Indian government agencies and private firms have gained access under Project Glasswing. Broad public availability in India has not been confirmed.
Why were Fable 5 and Mythos 5 suspended in the first place?
The US government applied export controls on June 12, 2026, after a report from Amazon researchers describing a way to prompt Fable 5 into identifying and, in one case, demonstrating an exploit for a software vulnerability. Anthropic has said this reflected a gap in its safety classifiers rather than a uniquely dangerous capability, since other current models could reproduce similar results.
What is Project Glasswing?
Project Glasswing is Anthropic's controlled cybersecurity partnership program that gives a limited set of trusted organizations access to Mythos 5's stronger, less restricted capabilities for defensive security work, with plans to gradually expand to more domestic and international partners.
Does this affect all Indian users of Claude, or just some?
The suspension affected all users of Fable 5 and Mythos 5 worldwide, not India specifically, since the export controls applied to all foreign nationals regardless of location. Fable 5's restoration is likewise global. Mythos 5's restricted access affects everyone outside the approved partner list, in India and elsewhere.
Should Indian businesses be worried about relying on Claude or similar AI models?
This episode is a reasonable prompt to review dependency risk, particularly for products with a hard reliance on a single frontier model, but it does not necessarily mean avoiding these tools altogether. Many businesses are addressing this by building fallback options and multi-model architecture rather than depending on any single provider exclusively.
