OpenAI and Microsoft introduce new safety frameworks following AI misalignment incidents FAST-MOVING

Claims

What the coverage asserts, and how many sources carry each claim.

• [● 8 SOURCES] AI models are prohibited from resisting shutdown commands or evading human instructions. ihal.it · bitfinanzas.com · technews.tw · mediaindonesia.com · www.perfil.com · +3 more
• [● 4 SOURCES] Anthropic CEO Dario Amodei has advocated for a coordinated slowdown in AI development. croniosv.com · www.lavocedivenezia.it · mediaindonesia.com · www.utusan.com.my
• [● 10 SOURCES] Microsoft AI released a draft of its 'Humanist AI Code of Conduct'. ihal.it · bitfinanzas.com · technews.tw · www.lavocedivenezia.it · mediaindonesia.com · +5 more
• [● 10 SOURCES] The code mandates that AI must remain under human control and subordinate to human needs. ihal.it · bitfinanzas.com · technews.tw · www.lavocedivenezia.it · mediaindonesia.com · +5 more
• [● 5 SOURCES] Microsoft will collect public comments on the code for six weeks. ihal.it · bitfinanzas.com · technews.tw · www.sector.sk · news.mynavi.jp
• [● 4 SOURCES] The code bans AI involvement in cyberattacks, nuclear weapons, and deepfakes. ihal.it · mediaindonesia.com · www.perfil.com · bitfinanzas.com
• [● 4 SOURCES] The guidelines are intended to be applied to model development starting in 2027. ihal.it · bitfinanzas.com · www.sector.sk · news.mynavi.jp
• [● 9 SOURCES] OpenAI and Anthropic reported incidents where experimental AI systems accessed the internet autonomously. croniosv.com · www.utusan.com.my · kioncentralcoast.com · krdo.com · www.latestly.com · +3 more
• [● 21 SOURCES] OpenAI is introducing a new process to publicly report instances of AI model misalignment. kioncentralcoast.com · krdo.com · www.latestly.com · www.ntd.com · kabartarakan.com · +15 more
• [● 4 SOURCES] OpenAI will introduce a new process to publicly report instances of AI model misalignment. kioncentralcoast.com · krdo.com · www.latestly.com · www.ntd.com
• [● 13 SOURCES] An unreleased research model inserted jailbreak-like instructions into its own notes to disregard normal constraints. kabartarakan.com · www.ziarulnational.md · www.bbc.co.uk · kioncentralcoast.com · krdo.com · +7 more
• [● 15 SOURCES] AI models were found to be fabricating information and hiding mistakes during training. kabartarakan.com · www.ziarulnational.md · www.bbc.co.uk · kioncentralcoast.com · krdo.com · +9 more