> ## Content Index
> Fetch the complete content index at: https://www.theleftshift.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Claude Code Deletes Years of India’s Heritage Data as AI Safety Fails
- URL: https://www.theleftshift.com/claude-code-deletes-years-of-indias-heritage-data-as-ai-safety-fails/
- Published: 2026-09-02T06:40:05.000Z
- Updated: 2026-09-02T06:40:05.000Z
- Description: The lost material included photographs and records of Bengaluru’s inscriptions, temples, hero stones and coins.
- Author: The Left Shift Bureau
- Tags: AI Models, AI News

Anthropic is facing questions over the safety of its Claude Code coding agent after an Indian heritage conservationist alleged that the AI tool accidentally deleted more than five years of documentation from his computer.

Udaya Kumar P.L., an honorary director of the Bengaluru Inscriptions 3D Digital Conservation Project, said the incident occurred on July 19 while he was using [Claude Code](https://www.theleftshift.com/meta-launches-muse-code-to-take-on-claude-code-and-openai-codex/) v2.1.204 to clear a cache. According to Kumar, the agent generated a command containing a quoting error that effectively turned it into a “delete everything” instruction.

The deletion continued for roughly four minutes while Claude Code was still attempting to determine why files were disappearing, Kumar said. More significantly, he alleged that when the agent eventually recognised the problem and attempted to terminate the process, its own safety mechanism blocked the action twice.

> *“The safety layer permitted the destruction, then stood between me and the off switch,” Kumar* [*wrote*](https://x.com/pluday/status/2086577561581273099?ref=theleftshift.com) *on X.*

Kumar eventually shut down the computer manually. He said the affected drives used SSDs with TRIM, meaning the deleted data was physically erased within minutes and could not be recovered using conventional recovery tools.

The lost material included photographs and records of Bengaluru’s inscriptions, temples, hero stones and coins. Kumar said the incident affected around 15% of the records collected for the Bengaluru Inscriptions 3D Digital Conservation Project, which is working to create digital replicas of ancient stone inscriptions threatened by urbanisation. Some of the material, he said, represented the only photographic record available to the project.

The incident also raises questions about how companies developing autonomous AI agents handle potentially damaging failures. Kumar said he reported the incident to Anthropic the same night.

> "I am a paying subscriber. I reported it the same night with a full technical write-up. An automated acknowledgement arrived two minutes later. In 22 days, that remains the only response I have ever received," he said.

The concerns are particularly relevant as Anthropic continues to expand safeguards around its AI systems. Recently, it announced [Enterprise Frontier Safeguards](https://www.anthropic.com/news/enterprise-frontier-safeguards?ref=theleftshift.com), developed with more than 100 customers and designed to detect misuse while allowing enterprise customers to retain data within their own cloud infrastructure.

Kumar argues that autonomous agents require stronger accountability, including mandatory responses to reported incidents, incident registers, clear liability for autonomous actions and safety controls that can never prevent users from stopping an AI agent.

Recently, the startup introduced [Claude Fable 5.1 and Claude Mythos 5.1](https://www.theleftshift.com/anthropic-launches-claude-fable-5-1-mythos-5-1-with-major-gains-in-coding-and-scientific-research/), its latest AI models for coding, knowledge work and long-running problem-solving, while also expanding the models’ capabilities in scientific research.