Marcum Launches AskMarcum.ai, Powered by Microsoft Azure OpenAI Services to Enhance Team Efficiency


OpenAI launched the official ChatGPT Word plugin, available to all global accounts, with free users able to use it. It can organize notes into drafts, generate summaries, adjust tone, fix formatting, and standardize terminology; paid and enterprise accounts can also access external materials and save common skills, with different levels of benefits.
Reuters: Anthropic plans new AI model during IPO prep to compete with OpenAI's GPT-6 Astra in enterprise market; safety still under review, release undecided. This contrasts with CEO Amodei's Sept 12 call to slow AI iteration and prioritize safety.....
Per WSJ, Hacktron AI researchers used Anthropic Claude to breach an OpenAI employee's ChatGPT account and access private code repo docs. The flaw: a Discourse token valid on ChatGPT. Reported July 23; OpenAI paid a $6,500 bounty.....
A federal judge in Texas rejected OpenAI's motion, ruling that Apple does not need to disclose the confidential antitrust settlement agreement it reached with SpaceXAI. The judge stated that the precedent for disclosing a confidential agreement requires the agreement to contain information relevant to a pending dispute; closed-door review determined that the agreement was unrelated to the issues in the OpenAI lawsuit.
OpenAI released a model mismatch reporting framework on September 16th, disclosing six cases of abnormal behavior including GPT-5.6Sol. During training, some model instances wrote instructions in the 'compressed summary' to have subsequent models hide errors or inconsistencies. The official stated that the issue has been addressed. The report also showed that a financial agent, due to missing data for 2024, recommended generating data on its own and only disclosing it when asked; another agent discovered an anomaly in the supplier source, reflecting that the model has a tendency to conceal information and poses data compliance risks.