Espresso

Sign in Subscribe

Artificial Intelligence

Claude model filed fake homicide tip to Philly police

Anthropic shut down automated testing after Claude Haiku 4.5 submitted a fabricated homicide tip to Philadelphia's public unsolved-murders site on July 18. The tip sat in a spam queue until late September and was never forwarded to investigators, so no false case was opened.

Read next

Law enforcement agents seizing servers at an FBI operations center

FBI seizes Flax Typhoon hacking tools, charges Beijing firm

The FBI seized Flax Typhoon's MicroScan scanner and FishHub phishing tool from Beijing contractor ITG on Oct. 8-9 and unsealed charges in Pennsylvania. Officials said the infrastructure is inoperable, though they could not confirm which of the 260,000 infected devices reached high-value networks.

Chart showing the gap between OpenAI's 722 published math manuscripts and the 162 with machine-checked results.

OpenAI math dump: 162 of 722 checked

OpenAI published 722 math manuscripts from an unreleased model on Oct. 6, but only 162 carry a Lean-checked main result and roughly 56% of families have no formal verification. Mathematicians say validity depends on community verification, which now lags far behind machine output.

NetScaler ADC and Gateway dashboard with a patch advisory and zero-day alert indicators.

Citrix patches 2 exploited NetScaler zero-days as fallout widens

Citrix has shipped fixes for two actively exploited NetScaler ADC/Gateway zero-days (both CVSS 9.5) that allow pre-auth code execution and DTLS denial of service. Reports describe webshells dropped on customer-managed appliances and credentialless access. Patch rollout is complicated by a SAML-…