Our method: distinguishing facts, citing sources and calmly explaining what can really change for creators, independent and digital professionals.
IA
Anthropic and Accenture commit at least 2 billion dollars in integrated evaluation of IA models
On 18 September, Anthropic announced a partnership with Accenture, led by its specialist subsidiary Faculty, to evaluate and test its frontier models. Each company plans to invest at least a billion dollars over five years to develop this capability. Embedded evaluators will have access comparable to employees to follow training, review deployment decisions, test safeguards and report incidents. Anthropic notes, however, that the arrangement is still being developed: rules for access, publication of results and independent funding are not yet standardised.
What it can change: To choose a professional AI tool, add audits, incident reports and transparency on safeguards to your usual price and performance criteria. A significant financial commitment does not yet replace verifiable public results.
Read the source: Anthropic →
Digital
Hacktron documents a chain of HEIF faults that allowed access to internal accounts and deposits from OpenAI
In a technical report released on September 13, Hacktron explains that it combined a libheif decoder flaw used to process images and a poor authentication configuration to compromise accounts related to the OpenAI forum. Researchers report that they have reported problems responsibly and obtained confirmation of an OpenAI correction on 25 July; Speech also prepared a corrective and reinforced insulation of the image processing. The report recommends updating libheif and libde265, deactivate HEIF or AVIF formats when they are not necessary, and dissociate the decoding of unreliable files.
What it can change: If your site, forum or application accepts images, update system dependencies and containers, not just the administration interface. Isolate file processing, limit administrator accounts and enable multifactor authentication.
Read source: Hacktron AI →