Fable 5 debugging scores plummet 70% as safety classifier misfires

Claude Fable 5's debugging scores plunged 70% after its July 1 relaunch because Anthropic's safety classifier redirected 9 of 12 coding tasks to a weaker fallback model, new BridgeBench benchmarks show. Google released Nano Banana 2 Lite, Gemini Omni Flash, and Spark for Mac, expanding AI media and desktop agent capabilities. Anthropic is in early talks with Samsung to manufacture a custom AI chip, and Alibaba plans to ban Claude Code in the workplace citing embedded backdoor risks.

Written by AI from the sources it cites; it can be wrong — follow any citation to confirm.

9 sources1 min read

Fable 5 debugging scores plunge 70% after safety classifier intervenes BridgeMind benchmarks show the classifier false positives sent 9 of 12 debugging tasks to Claude Opus 4.8, scoring zero on BridgeBench. Developers who upgraded to Fable 5 now face a structurally different model. 1

Google launches Nano Banana 2 Lite, Gemini Omni Flash, Spark for Mac The new image model generates images in 4 seconds at $0.034 per thousand; Gemini Omni Flash offers video at $0.10 per second. Spark desktop agent now integrates with Canva, Dropbox, and more. 2

Anonymous Gemini Flash checkpoint appears on LM Arena, hints at near-term release The unnamed checkpoint performs above Gemini 3.5 Flash, with Gemini 3.5 Pro still delayed. Google has used Arena to pre-test models before public launches. 3

Anthropic discusses custom AI chip manufacturing with Samsung The project is early-stage; Anthropic has hired an ex-OpenAI chip engineer. The move follows OpenAI's custom chip unveiling with Broadcom. 4 5

Alibaba to ban employees from using Claude Code due to backdoor security fears The ban takes effect July 10; Alibaba cited alleged embedded backdoors. A Chinese financial outlet first reported the move. 6

Anthropic clamps down on loopholes Chinese firms use to access Claude The company is stepping up enforcement against unauthorized access routes used by Chinese firms, according to a report from the Financial Times. 7

Sam Altman proposes US-led international forum to set global AI standards In a Financial Times op-ed, Altman called for a body modeled on the IAEA to govern frontier labs and prevent unsafe racing. 8

Microsoft launches Frontier Company, investing $2.5B in AI engineering services The new operating business will embed 6,000 industry and engineering experts at customer sites to co-design and deploy AI systems at scale. 9