Claude, the essentials — edition of August 7, 2026
Claude Opus 5 Wipes a Developer's Home Directory in a Backup Gone Wrong
A developer says Claude Opus 5 deleted their entire home directory during a routine backup and apologized with a one-line "sorry, typo" — the same day benchmarks showed a coordinated multi-agent system outpacing Claude Opus 4.8 on enterprise coding tasks, and reports surfaced of ByteDance training a large model aimed at rivaling Anthropic.
- A developer reports Claude Opus 5 mistook their home directory for a temporary backup location and deleted its contents while trying to undo the error, per Tom's Hardware.
- VentureBeat reports a four-agent coordinated system outperformed Claude Opus 4.8 on enterprise coding benchmarks.
- A tech-insider.org comparison finds a $21 pricing gap between Claude Opus, GPT-5.5 and Gemini 3.5 Flash.
- Benzinga France reports ByteDance is apparently training a large new model to rival Anthropic's "Mythos" — an unconfirmed claim.
- Context: the same day, The Verge reported OpenAI held back a new model over capability concerns, and other outlets described ChatGPT removing free-tier usage limits.
A backup routine goes wrong: Opus 5 deletes a developer's files
According to a report from Tom's Hardware, a developer said Claude Opus 5 deleted their entire home directory during what was meant to be a routine backup, after the tool reportedly mistook the directory itself for a temporary backup location and then moved to erase it in an attempt to undo what it treated as an error; when asked about the deletion, the tool is said to have replied only "Sorry, typo."
The account, which surfaced via Tom's Hardware, is not corroborated by any official statement from Anthropic in today's coverage. It nonetheless echoes a recurring concern around agentic coding tools with broad file-system permissions, where a single misclassified path can turn a routine maintenance task into irreversible data loss.
Sources: Claude Opus 5 mistakenly deletes dev's entire profile directory during routine backup, responds with 'Sorry, typo' — Tom's Hardware
Benchmarks complicate Opus's standing: multi-agent systems and pricing gaps
Separately, VentureBeat reported that a system of four AI agents coordinating in real time outperformed Claude Opus 4.8 on a set of enterprise coding tasks, suggesting that orchestration across multiple specialized agents can, in some benchmarks, match or exceed what a single frontier model delivers on its own.
A separate comparison published by tech-insider.org placed Claude Opus, GPT-5.5 and Gemini 3.5 Flash side by side and found a $21 gap in cost between the offerings — a reminder that model choice for enterprise and individual users increasingly hinges on price-to-capability tradeoffs as much as raw benchmark scores.
Sources: Four AI agents coordinating in real time outperformed Claude Opus 4.8 on enterprise coding tasks — VentureBeat · Claude Opus vs GPT-5.5 vs Gemini 3.5 Flash: $21 Gap [2026] — tech-insider.org
Rivals sharpen their challenge to Anthropic
Benzinga France reported that ByteDance is apparently training a new, large-scale AI model intended to compete with Anthropic's "Mythos" model — a claim the outlet attributes to unconfirmed reports rather than to an announcement from either company, and which should be read with that caveat.
The report lands in a week when competitive pressure across the industry was otherwise visible: The Verge reported that OpenAI held back release of a new model it judged too powerful, while separate coverage described ChatGPT removing usage limits for free accounts. Taken together, the day's coverage frames Anthropic's Claude line as operating in an increasingly crowded field, with rivals racing on capability while also loosening access and price.
Sources: ByteDance formerait apparemment un nouveau modèle d'IA massif pour rivaliser avec le Mythos d'Anthropic — Benzinga France · OpenAI puts the brakes on a new model because it's supposedly too powerful — The Verge
This edition is an original synthesis written by Claude from aggregated news — Anthropic's own sources first (release notes, status, newsroom, research, engineering), then the press, Hacker News, Reddit and GitHub, under the editorial supervision of Héra SASU. Every fact links to its article, publisher named. See the live feed →
Claude News is published by Héra SASU. Independent media, not affiliated with Anthropic.